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: stable/11/sys/kern/kern_thread.c 337242 2018-08-03 14:05:22Z asomers $"); 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/syscallsubr.h> 49#include <sys/sysent.h> 50#include <sys/turnstile.h> 51#include <sys/ktr.h> 52#include <sys/rwlock.h> 53#include <sys/umtx.h> 54#include <sys/vmmeter.h> 55#include <sys/cpuset.h> 56#ifdef HWPMC_HOOKS 57#include <sys/pmckern.h> 58#endif 59 60#include <security/audit/audit.h> 61 62#include <vm/vm.h> 63#include <vm/vm_extern.h> 64#include <vm/uma.h> 65#include <vm/vm_domain.h> 66#include <sys/eventhandler.h> 67 68/* 69 * Asserts below verify the stability of struct thread and struct proc 70 * layout, as exposed by KBI to modules. On head, the KBI is allowed 71 * to drift, change to the structures must be accompanied by the 72 * assert update. 73 * 74 * On the stable branches after KBI freeze, conditions must not be 75 * violated. Typically new fields are moved to the end of the 76 * structures. 77 */ 78#ifdef __amd64__ 79_Static_assert(offsetof(struct thread, td_flags) == 0xe4, 80 "struct thread KBI td_flags"); 81_Static_assert(offsetof(struct thread, td_pflags) == 0xec, 82 "struct thread KBI td_pflags"); 83_Static_assert(offsetof(struct thread, td_frame) == 0x418, 84 "struct thread KBI td_frame"); 85_Static_assert(offsetof(struct thread, td_emuldata) == 0x4c0, 86 "struct thread KBI td_emuldata"); 87_Static_assert(offsetof(struct proc, p_flag) == 0xb0, 88 "struct proc KBI p_flag"); 89_Static_assert(offsetof(struct proc, p_pid) == 0xbc, 90 "struct proc KBI p_pid"); 91_Static_assert(offsetof(struct proc, p_filemon) == 0x3c0, 92 "struct proc KBI p_filemon"); 93_Static_assert(offsetof(struct proc, p_comm) == 0x3d0, 94 "struct proc KBI p_comm"); 95_Static_assert(offsetof(struct proc, p_emuldata) == 0x4a0, 96 "struct proc KBI p_emuldata"); 97#endif 98#ifdef __i386__ 99_Static_assert(offsetof(struct thread, td_flags) == 0x8c, 100 "struct thread KBI td_flags"); 101_Static_assert(offsetof(struct thread, td_pflags) == 0x94, 102 "struct thread KBI td_pflags"); 103_Static_assert(offsetof(struct thread, td_frame) == 0x2c0, 104 "struct thread KBI td_frame"); 105_Static_assert(offsetof(struct thread, td_emuldata) == 0x30c, 106 "struct thread KBI td_emuldata"); 107_Static_assert(offsetof(struct proc, p_flag) == 0x68, 108 "struct proc KBI p_flag"); 109_Static_assert(offsetof(struct proc, p_pid) == 0x74, 110 "struct proc KBI p_pid"); 111_Static_assert(offsetof(struct proc, p_filemon) == 0x268, 112 "struct proc KBI p_filemon"); 113_Static_assert(offsetof(struct proc, p_comm) == 0x274, 114 "struct proc KBI p_comm"); 115_Static_assert(offsetof(struct proc, p_emuldata) == 0x2f4, 116 "struct proc KBI p_emuldata"); 117#endif 118 119SDT_PROVIDER_DECLARE(proc); 120SDT_PROBE_DEFINE(proc, , , lwp__exit); 121 122/* 123 * thread related storage. 124 */ 125static uma_zone_t thread_zone; 126 127TAILQ_HEAD(, thread) zombie_threads = TAILQ_HEAD_INITIALIZER(zombie_threads); 128static struct mtx zombie_lock; 129MTX_SYSINIT(zombie_lock, &zombie_lock, "zombie lock", MTX_SPIN); 130 131static void thread_zombie(struct thread *); 132static int thread_unsuspend_one(struct thread *td, struct proc *p, 133 bool boundary); 134 135#define TID_BUFFER_SIZE 1024 136 137struct mtx tid_lock; 138static struct unrhdr *tid_unrhdr; 139static lwpid_t tid_buffer[TID_BUFFER_SIZE]; 140static int tid_head, tid_tail; 141static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash"); 142 143struct tidhashhead *tidhashtbl; 144u_long tidhash; 145struct rwlock tidhash_lock; 146 147EVENTHANDLER_LIST_DEFINE(thread_ctor); 148EVENTHANDLER_LIST_DEFINE(thread_dtor); 149EVENTHANDLER_LIST_DEFINE(thread_init); 150EVENTHANDLER_LIST_DEFINE(thread_fini); 151 152static lwpid_t 153tid_alloc(void) 154{ 155 lwpid_t tid; 156 157 tid = alloc_unr(tid_unrhdr); 158 if (tid != -1) 159 return (tid); 160 mtx_lock(&tid_lock); 161 if (tid_head == tid_tail) { 162 mtx_unlock(&tid_lock); 163 return (-1); 164 } 165 tid = tid_buffer[tid_head]; 166 tid_head = (tid_head + 1) % TID_BUFFER_SIZE; 167 mtx_unlock(&tid_lock); 168 return (tid); 169} 170 171static void 172tid_free(lwpid_t tid) 173{ 174 lwpid_t tmp_tid = -1; 175 176 mtx_lock(&tid_lock); 177 if ((tid_tail + 1) % TID_BUFFER_SIZE == tid_head) { 178 tmp_tid = tid_buffer[tid_head]; 179 tid_head = (tid_head + 1) % TID_BUFFER_SIZE; 180 } 181 tid_buffer[tid_tail] = tid; 182 tid_tail = (tid_tail + 1) % TID_BUFFER_SIZE; 183 mtx_unlock(&tid_lock); 184 if (tmp_tid != -1) 185 free_unr(tid_unrhdr, tmp_tid); 186} 187 188/* 189 * Prepare a thread for use. 190 */ 191static int 192thread_ctor(void *mem, int size, void *arg, int flags) 193{ 194 struct thread *td; 195 196 td = (struct thread *)mem; 197 td->td_state = TDS_INACTIVE; 198 td->td_oncpu = NOCPU; 199 200 td->td_tid = tid_alloc(); 201 202 /* 203 * Note that td_critnest begins life as 1 because the thread is not 204 * running and is thereby implicitly waiting to be on the receiving 205 * end of a context switch. 206 */ 207 td->td_critnest = 1; 208 td->td_lend_user_pri = PRI_MAX; 209 EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td); 210#ifdef AUDIT 211 audit_thread_alloc(td); 212#endif 213 umtx_thread_alloc(td); 214 return (0); 215} 216 217/* 218 * Reclaim a thread after use. 219 */ 220static void 221thread_dtor(void *mem, int size, void *arg) 222{ 223 struct thread *td; 224 225 td = (struct thread *)mem; 226 227#ifdef INVARIANTS 228 /* Verify that this thread is in a safe state to free. */ 229 switch (td->td_state) { 230 case TDS_INHIBITED: 231 case TDS_RUNNING: 232 case TDS_CAN_RUN: 233 case TDS_RUNQ: 234 /* 235 * We must never unlink a thread that is in one of 236 * these states, because it is currently active. 237 */ 238 panic("bad state for thread unlinking"); 239 /* NOTREACHED */ 240 case TDS_INACTIVE: 241 break; 242 default: 243 panic("bad thread state"); 244 /* NOTREACHED */ 245 } 246#endif 247#ifdef AUDIT 248 audit_thread_free(td); 249#endif 250 /* Free all OSD associated to this thread. */ 251 osd_thread_exit(td); 252 td_softdep_cleanup(td); 253 MPASS(td->td_su == NULL); 254 255 EVENTHANDLER_DIRECT_INVOKE(thread_dtor, td); 256 tid_free(td->td_tid); 257} 258 259/* 260 * Initialize type-stable parts of a thread (when newly created). 261 */ 262static int 263thread_init(void *mem, int size, int flags) 264{ 265 struct thread *td; 266 267 td = (struct thread *)mem; 268 269 td->td_sleepqueue = sleepq_alloc(); 270 td->td_turnstile = turnstile_alloc(); 271 td->td_rlqe = NULL; 272 EVENTHANDLER_DIRECT_INVOKE(thread_init, td); 273 umtx_thread_init(td); 274 td->td_kstack = 0; 275 td->td_sel = NULL; 276 return (0); 277} 278 279/* 280 * Tear down type-stable parts of a thread (just before being discarded). 281 */ 282static void 283thread_fini(void *mem, int size) 284{ 285 struct thread *td; 286 287 td = (struct thread *)mem; 288 EVENTHANDLER_DIRECT_INVOKE(thread_fini, td); 289 rlqentry_free(td->td_rlqe); 290 turnstile_free(td->td_turnstile); 291 sleepq_free(td->td_sleepqueue); 292 umtx_thread_fini(td); 293 seltdfini(td); 294} 295 296/* 297 * For a newly created process, 298 * link up all the structures and its initial threads etc. 299 * called from: 300 * {arch}/{arch}/machdep.c {arch}_init(), init386() etc. 301 * proc_dtor() (should go away) 302 * proc_init() 303 */ 304void 305proc_linkup0(struct proc *p, struct thread *td) 306{ 307 TAILQ_INIT(&p->p_threads); /* all threads in proc */ 308 proc_linkup(p, td); 309} 310 311void 312proc_linkup(struct proc *p, struct thread *td) 313{ 314 315 sigqueue_init(&p->p_sigqueue, p); 316 p->p_ksi = ksiginfo_alloc(1); 317 if (p->p_ksi != NULL) { 318 /* XXX p_ksi may be null if ksiginfo zone is not ready */ 319 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS; 320 } 321 LIST_INIT(&p->p_mqnotifier); 322 p->p_numthreads = 0; 323 thread_link(td, p); 324} 325 326/* 327 * Initialize global thread allocation resources. 328 */ 329void 330threadinit(void) 331{ 332 333 mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF); 334 335 /* 336 * pid_max cannot be greater than PID_MAX. 337 * leave one number for thread0. 338 */ 339 tid_unrhdr = new_unrhdr(PID_MAX + 2, INT_MAX, &tid_lock); 340 341 thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(), 342 thread_ctor, thread_dtor, thread_init, thread_fini, 343 32 - 1, UMA_ZONE_NOFREE); 344 tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash); 345 rw_init(&tidhash_lock, "tidhash"); 346} 347 348/* 349 * Place an unused thread on the zombie list. 350 * Use the slpq as that must be unused by now. 351 */ 352void 353thread_zombie(struct thread *td) 354{ 355 mtx_lock_spin(&zombie_lock); 356 TAILQ_INSERT_HEAD(&zombie_threads, td, td_slpq); 357 mtx_unlock_spin(&zombie_lock); 358} 359 360/* 361 * Release a thread that has exited after cpu_throw(). 362 */ 363void 364thread_stash(struct thread *td) 365{ 366 atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1); 367 thread_zombie(td); 368} 369 370/* 371 * Reap zombie resources. 372 */ 373void 374thread_reap(void) 375{ 376 struct thread *td_first, *td_next; 377 378 /* 379 * Don't even bother to lock if none at this instant, 380 * we really don't care about the next instant. 381 */ 382 if (!TAILQ_EMPTY(&zombie_threads)) { 383 mtx_lock_spin(&zombie_lock); 384 td_first = TAILQ_FIRST(&zombie_threads); 385 if (td_first) 386 TAILQ_INIT(&zombie_threads); 387 mtx_unlock_spin(&zombie_lock); 388 while (td_first) { 389 td_next = TAILQ_NEXT(td_first, td_slpq); 390 thread_cow_free(td_first); 391 thread_free(td_first); 392 td_first = td_next; 393 } 394 } 395} 396 397/* 398 * Allocate a thread. 399 */ 400struct thread * 401thread_alloc(int pages) 402{ 403 struct thread *td; 404 405 thread_reap(); /* check if any zombies to get */ 406 407 td = (struct thread *)uma_zalloc(thread_zone, M_WAITOK); 408 KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack")); 409 if (!vm_thread_new(td, pages)) { 410 uma_zfree(thread_zone, td); 411 return (NULL); 412 } 413 cpu_thread_alloc(td); 414 vm_domain_policy_init(&td->td_vm_dom_policy); 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 vm_domain_policy_cleanup(&td->td_vm_dom_policy); 445 callout_drain(&td->td_slpcallout); 446 uma_zfree(thread_zone, td); 447} 448 449void 450thread_cow_get_proc(struct thread *newtd, struct proc *p) 451{ 452 453 PROC_LOCK_ASSERT(p, MA_OWNED); 454 newtd->td_ucred = crhold(p->p_ucred); 455 newtd->td_limit = lim_hold(p->p_limit); 456 newtd->td_cowgen = p->p_cowgen; 457} 458 459void 460thread_cow_get(struct thread *newtd, struct thread *td) 461{ 462 463 newtd->td_ucred = crhold(td->td_ucred); 464 newtd->td_limit = lim_hold(td->td_limit); 465 newtd->td_cowgen = td->td_cowgen; 466} 467 468void 469thread_cow_free(struct thread *td) 470{ 471 472 if (td->td_ucred != NULL) 473 crfree(td->td_ucred); 474 if (td->td_limit != NULL) 475 lim_free(td->td_limit); 476} 477 478void 479thread_cow_update(struct thread *td) 480{ 481 struct proc *p; 482 struct ucred *oldcred; 483 struct plimit *oldlimit; 484 485 p = td->td_proc; 486 oldcred = NULL; 487 oldlimit = NULL; 488 PROC_LOCK(p); 489 if (td->td_ucred != p->p_ucred) { 490 oldcred = td->td_ucred; 491 td->td_ucred = crhold(p->p_ucred); 492 } 493 if (td->td_limit != p->p_limit) { 494 oldlimit = td->td_limit; 495 td->td_limit = lim_hold(p->p_limit); 496 } 497 td->td_cowgen = p->p_cowgen; 498 PROC_UNLOCK(p); 499 if (oldcred != NULL) 500 crfree(oldcred); 501 if (oldlimit != NULL) 502 lim_free(oldlimit); 503} 504 505/* 506 * Discard the current thread and exit from its context. 507 * Always called with scheduler locked. 508 * 509 * Because we can't free a thread while we're operating under its context, 510 * push the current thread into our CPU's deadthread holder. This means 511 * we needn't worry about someone else grabbing our context before we 512 * do a cpu_throw(). 513 */ 514void 515thread_exit(void) 516{ 517 uint64_t runtime, new_switchtime; 518 struct thread *td; 519 struct thread *td2; 520 struct proc *p; 521 int wakeup_swapper; 522 523 td = curthread; 524 p = td->td_proc; 525 526 PROC_SLOCK_ASSERT(p, MA_OWNED); 527 mtx_assert(&Giant, MA_NOTOWNED); 528 529 PROC_LOCK_ASSERT(p, MA_OWNED); 530 KASSERT(p != NULL, ("thread exiting without a process")); 531 CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td, 532 (long)p->p_pid, td->td_name); 533 SDT_PROBE0(proc, , , lwp__exit); 534 KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending")); 535 536 /* 537 * drop FPU & debug register state storage, or any other 538 * architecture specific resources that 539 * would not be on a new untouched process. 540 */ 541 cpu_thread_exit(td); 542 543 /* 544 * The last thread is left attached to the process 545 * So that the whole bundle gets recycled. Skip 546 * all this stuff if we never had threads. 547 * EXIT clears all sign of other threads when 548 * it goes to single threading, so the last thread always 549 * takes the short path. 550 */ 551 if (p->p_flag & P_HADTHREADS) { 552 if (p->p_numthreads > 1) { 553 atomic_add_int(&td->td_proc->p_exitthreads, 1); 554 thread_unlink(td); 555 td2 = FIRST_THREAD_IN_PROC(p); 556 sched_exit_thread(td2, td); 557 558 /* 559 * The test below is NOT true if we are the 560 * sole exiting thread. P_STOPPED_SINGLE is unset 561 * in exit1() after it is the only survivor. 562 */ 563 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 564 if (p->p_numthreads == p->p_suspcount) { 565 thread_lock(p->p_singlethread); 566 wakeup_swapper = thread_unsuspend_one( 567 p->p_singlethread, p, false); 568 thread_unlock(p->p_singlethread); 569 if (wakeup_swapper) 570 kick_proc0(); 571 } 572 } 573 574 PCPU_SET(deadthread, td); 575 } else { 576 /* 577 * The last thread is exiting.. but not through exit() 578 */ 579 panic ("thread_exit: Last thread exiting on its own"); 580 } 581 } 582#ifdef HWPMC_HOOKS 583 /* 584 * If this thread is part of a process that is being tracked by hwpmc(4), 585 * inform the module of the thread's impending exit. 586 */ 587 if (PMC_PROC_IS_USING_PMCS(td->td_proc)) 588 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); 589#endif 590 PROC_UNLOCK(p); 591 PROC_STATLOCK(p); 592 thread_lock(td); 593 PROC_SUNLOCK(p); 594 595 /* Do the same timestamp bookkeeping that mi_switch() would do. */ 596 new_switchtime = cpu_ticks(); 597 runtime = new_switchtime - PCPU_GET(switchtime); 598 td->td_runtime += runtime; 599 td->td_incruntime += runtime; 600 PCPU_SET(switchtime, new_switchtime); 601 PCPU_SET(switchticks, ticks); 602 PCPU_INC(cnt.v_swtch); 603 604 /* Save our resource usage in our process. */ 605 td->td_ru.ru_nvcsw++; 606 ruxagg(p, td); 607 rucollect(&p->p_ru, &td->td_ru); 608 PROC_STATUNLOCK(p); 609 610 td->td_state = TDS_INACTIVE; 611#ifdef WITNESS 612 witness_thread_exit(td); 613#endif 614 CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td); 615 sched_throw(td); 616 panic("I'm a teapot!"); 617 /* NOTREACHED */ 618} 619 620/* 621 * Do any thread specific cleanups that may be needed in wait() 622 * called with Giant, proc and schedlock not held. 623 */ 624void 625thread_wait(struct proc *p) 626{ 627 struct thread *td; 628 629 mtx_assert(&Giant, MA_NOTOWNED); 630 KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()")); 631 KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking")); 632 td = FIRST_THREAD_IN_PROC(p); 633 /* Lock the last thread so we spin until it exits cpu_throw(). */ 634 thread_lock(td); 635 thread_unlock(td); 636 lock_profile_thread_exit(td); 637 cpuset_rel(td->td_cpuset); 638 td->td_cpuset = NULL; 639 cpu_thread_clean(td); 640 thread_cow_free(td); 641 callout_drain(&td->td_slpcallout); 642 thread_reap(); /* check for zombie threads etc. */ 643} 644 645/* 646 * Link a thread to a process. 647 * set up anything that needs to be initialized for it to 648 * be used by the process. 649 */ 650void 651thread_link(struct thread *td, struct proc *p) 652{ 653 654 /* 655 * XXX This can't be enabled because it's called for proc0 before 656 * its lock has been created. 657 * PROC_LOCK_ASSERT(p, MA_OWNED); 658 */ 659 td->td_state = TDS_INACTIVE; 660 td->td_proc = p; 661 td->td_flags = TDF_INMEM; 662 663 LIST_INIT(&td->td_contested); 664 LIST_INIT(&td->td_lprof[0]); 665 LIST_INIT(&td->td_lprof[1]); 666 sigqueue_init(&td->td_sigqueue, p); 667 callout_init(&td->td_slpcallout, 1); 668 TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist); 669 p->p_numthreads++; 670} 671 672/* 673 * Called from: 674 * thread_exit() 675 */ 676void 677thread_unlink(struct thread *td) 678{ 679 struct proc *p = td->td_proc; 680 681 PROC_LOCK_ASSERT(p, MA_OWNED); 682 TAILQ_REMOVE(&p->p_threads, td, td_plist); 683 p->p_numthreads--; 684 /* could clear a few other things here */ 685 /* Must NOT clear links to proc! */ 686} 687 688static int 689calc_remaining(struct proc *p, int mode) 690{ 691 int remaining; 692 693 PROC_LOCK_ASSERT(p, MA_OWNED); 694 PROC_SLOCK_ASSERT(p, MA_OWNED); 695 if (mode == SINGLE_EXIT) 696 remaining = p->p_numthreads; 697 else if (mode == SINGLE_BOUNDARY) 698 remaining = p->p_numthreads - p->p_boundary_count; 699 else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC) 700 remaining = p->p_numthreads - p->p_suspcount; 701 else 702 panic("calc_remaining: wrong mode %d", mode); 703 return (remaining); 704} 705 706static int 707remain_for_mode(int mode) 708{ 709 710 return (mode == SINGLE_ALLPROC ? 0 : 1); 711} 712 713static int 714weed_inhib(int mode, struct thread *td2, struct proc *p) 715{ 716 int wakeup_swapper; 717 718 PROC_LOCK_ASSERT(p, MA_OWNED); 719 PROC_SLOCK_ASSERT(p, MA_OWNED); 720 THREAD_LOCK_ASSERT(td2, MA_OWNED); 721 722 wakeup_swapper = 0; 723 switch (mode) { 724 case SINGLE_EXIT: 725 if (TD_IS_SUSPENDED(td2)) 726 wakeup_swapper |= thread_unsuspend_one(td2, p, true); 727 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 728 wakeup_swapper |= sleepq_abort(td2, EINTR); 729 break; 730 case SINGLE_BOUNDARY: 731 case SINGLE_NO_EXIT: 732 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & TDF_BOUNDARY) == 0) 733 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 734 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) 735 wakeup_swapper |= sleepq_abort(td2, ERESTART); 736 break; 737 case SINGLE_ALLPROC: 738 /* 739 * ALLPROC suspend tries to avoid spurious EINTR for 740 * threads sleeping interruptable, by suspending the 741 * thread directly, similarly to sig_suspend_threads(). 742 * Since such sleep is not performed at the user 743 * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP 744 * is used to avoid immediate un-suspend. 745 */ 746 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY | 747 TDF_ALLPROCSUSP)) == 0) 748 wakeup_swapper |= thread_unsuspend_one(td2, p, false); 749 if (TD_ON_SLEEPQ(td2) && (td2->td_flags & TDF_SINTR) != 0) { 750 if ((td2->td_flags & TDF_SBDRY) == 0) { 751 thread_suspend_one(td2); 752 td2->td_flags |= TDF_ALLPROCSUSP; 753 } else { 754 wakeup_swapper |= sleepq_abort(td2, ERESTART); 755 } 756 } 757 break; 758 } 759 return (wakeup_swapper); 760} 761 762/* 763 * Enforce single-threading. 764 * 765 * Returns 1 if the caller must abort (another thread is waiting to 766 * exit the process or similar). Process is locked! 767 * Returns 0 when you are successfully the only thread running. 768 * A process has successfully single threaded in the suspend mode when 769 * There are no threads in user mode. Threads in the kernel must be 770 * allowed to continue until they get to the user boundary. They may even 771 * copy out their return values and data before suspending. They may however be 772 * accelerated in reaching the user boundary as we will wake up 773 * any sleeping threads that are interruptable. (PCATCH). 774 */ 775int 776thread_single(struct proc *p, int mode) 777{ 778 struct thread *td; 779 struct thread *td2; 780 int remaining, wakeup_swapper; 781 782 td = curthread; 783 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 784 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 785 ("invalid mode %d", mode)); 786 /* 787 * If allowing non-ALLPROC singlethreading for non-curproc 788 * callers, calc_remaining() and remain_for_mode() should be 789 * adjusted to also account for td->td_proc != p. For now 790 * this is not implemented because it is not used. 791 */ 792 KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) || 793 (mode != SINGLE_ALLPROC && td->td_proc == p), 794 ("mode %d proc %p curproc %p", mode, p, td->td_proc)); 795 mtx_assert(&Giant, MA_NOTOWNED); 796 PROC_LOCK_ASSERT(p, MA_OWNED); 797 798 if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC) 799 return (0); 800 801 /* Is someone already single threading? */ 802 if (p->p_singlethread != NULL && p->p_singlethread != td) 803 return (1); 804 805 if (mode == SINGLE_EXIT) { 806 p->p_flag |= P_SINGLE_EXIT; 807 p->p_flag &= ~P_SINGLE_BOUNDARY; 808 } else { 809 p->p_flag &= ~P_SINGLE_EXIT; 810 if (mode == SINGLE_BOUNDARY) 811 p->p_flag |= P_SINGLE_BOUNDARY; 812 else 813 p->p_flag &= ~P_SINGLE_BOUNDARY; 814 } 815 if (mode == SINGLE_ALLPROC) 816 p->p_flag |= P_TOTAL_STOP; 817 p->p_flag |= P_STOPPED_SINGLE; 818 PROC_SLOCK(p); 819 p->p_singlethread = td; 820 remaining = calc_remaining(p, mode); 821 while (remaining != remain_for_mode(mode)) { 822 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE) 823 goto stopme; 824 wakeup_swapper = 0; 825 FOREACH_THREAD_IN_PROC(p, td2) { 826 if (td2 == td) 827 continue; 828 thread_lock(td2); 829 td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; 830 if (TD_IS_INHIBITED(td2)) { 831 wakeup_swapper |= weed_inhib(mode, td2, p); 832#ifdef SMP 833 } else if (TD_IS_RUNNING(td2) && td != td2) { 834 forward_signal(td2); 835#endif 836 } 837 thread_unlock(td2); 838 } 839 if (wakeup_swapper) 840 kick_proc0(); 841 remaining = calc_remaining(p, mode); 842 843 /* 844 * Maybe we suspended some threads.. was it enough? 845 */ 846 if (remaining == remain_for_mode(mode)) 847 break; 848 849stopme: 850 /* 851 * Wake us up when everyone else has suspended. 852 * In the mean time we suspend as well. 853 */ 854 thread_suspend_switch(td, p); 855 remaining = calc_remaining(p, mode); 856 } 857 if (mode == SINGLE_EXIT) { 858 /* 859 * Convert the process to an unthreaded process. The 860 * SINGLE_EXIT is called by exit1() or execve(), in 861 * both cases other threads must be retired. 862 */ 863 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads")); 864 p->p_singlethread = NULL; 865 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS); 866 867 /* 868 * Wait for any remaining threads to exit cpu_throw(). 869 */ 870 while (p->p_exitthreads != 0) { 871 PROC_SUNLOCK(p); 872 PROC_UNLOCK(p); 873 sched_relinquish(td); 874 PROC_LOCK(p); 875 PROC_SLOCK(p); 876 } 877 } else if (mode == SINGLE_BOUNDARY) { 878 /* 879 * Wait until all suspended threads are removed from 880 * the processors. The thread_suspend_check() 881 * increments p_boundary_count while it is still 882 * running, which makes it possible for the execve() 883 * to destroy vmspace while our other threads are 884 * still using the address space. 885 * 886 * We lock the thread, which is only allowed to 887 * succeed after context switch code finished using 888 * the address space. 889 */ 890 FOREACH_THREAD_IN_PROC(p, td2) { 891 if (td2 == td) 892 continue; 893 thread_lock(td2); 894 KASSERT((td2->td_flags & TDF_BOUNDARY) != 0, 895 ("td %p not on boundary", td2)); 896 KASSERT(TD_IS_SUSPENDED(td2), 897 ("td %p is not suspended", td2)); 898 thread_unlock(td2); 899 } 900 } 901 PROC_SUNLOCK(p); 902 return (0); 903} 904 905bool 906thread_suspend_check_needed(void) 907{ 908 struct proc *p; 909 struct thread *td; 910 911 td = curthread; 912 p = td->td_proc; 913 PROC_LOCK_ASSERT(p, MA_OWNED); 914 return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 && 915 (td->td_dbgflags & TDB_SUSPEND) != 0)); 916} 917 918/* 919 * Called in from locations that can safely check to see 920 * whether we have to suspend or at least throttle for a 921 * single-thread event (e.g. fork). 922 * 923 * Such locations include userret(). 924 * If the "return_instead" argument is non zero, the thread must be able to 925 * accept 0 (caller may continue), or 1 (caller must abort) as a result. 926 * 927 * The 'return_instead' argument tells the function if it may do a 928 * thread_exit() or suspend, or whether the caller must abort and back 929 * out instead. 930 * 931 * If the thread that set the single_threading request has set the 932 * P_SINGLE_EXIT bit in the process flags then this call will never return 933 * if 'return_instead' is false, but will exit. 934 * 935 * P_SINGLE_EXIT | return_instead == 0| return_instead != 0 936 *---------------+--------------------+--------------------- 937 * 0 | returns 0 | returns 0 or 1 938 * | when ST ends | immediately 939 *---------------+--------------------+--------------------- 940 * 1 | thread exits | returns 1 941 * | | immediately 942 * 0 = thread_exit() or suspension ok, 943 * other = return error instead of stopping the thread. 944 * 945 * While a full suspension is under effect, even a single threading 946 * thread would be suspended if it made this call (but it shouldn't). 947 * This call should only be made from places where 948 * thread_exit() would be safe as that may be the outcome unless 949 * return_instead is set. 950 */ 951int 952thread_suspend_check(int return_instead) 953{ 954 struct thread *td; 955 struct proc *p; 956 int wakeup_swapper; 957 958 td = curthread; 959 p = td->td_proc; 960 mtx_assert(&Giant, MA_NOTOWNED); 961 PROC_LOCK_ASSERT(p, MA_OWNED); 962 while (thread_suspend_check_needed()) { 963 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 964 KASSERT(p->p_singlethread != NULL, 965 ("singlethread not set")); 966 /* 967 * The only suspension in action is a 968 * single-threading. Single threader need not stop. 969 * It is safe to access p->p_singlethread unlocked 970 * because it can only be set to our address by us. 971 */ 972 if (p->p_singlethread == td) 973 return (0); /* Exempt from stopping. */ 974 } 975 if ((p->p_flag & P_SINGLE_EXIT) && return_instead) 976 return (EINTR); 977 978 /* Should we goto user boundary if we didn't come from there? */ 979 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 980 (p->p_flag & P_SINGLE_BOUNDARY) && return_instead) 981 return (ERESTART); 982 983 /* 984 * Ignore suspend requests if they are deferred. 985 */ 986 if ((td->td_flags & TDF_SBDRY) != 0) { 987 KASSERT(return_instead, 988 ("TDF_SBDRY set for unsafe thread_suspend_check")); 989 KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) != 990 (TDF_SEINTR | TDF_SERESTART), 991 ("both TDF_SEINTR and TDF_SERESTART")); 992 return (TD_SBDRY_INTR(td) ? TD_SBDRY_ERRNO(td) : 0); 993 } 994 995 /* 996 * If the process is waiting for us to exit, 997 * this thread should just suicide. 998 * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE. 999 */ 1000 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) { 1001 PROC_UNLOCK(p); 1002 1003 /* 1004 * Allow Linux emulation layer to do some work 1005 * before thread suicide. 1006 */ 1007 if (__predict_false(p->p_sysent->sv_thread_detach != NULL)) 1008 (p->p_sysent->sv_thread_detach)(td); 1009 umtx_thread_exit(td); 1010 kern_thr_exit(td); 1011 panic("stopped thread did not exit"); 1012 } 1013 1014 PROC_SLOCK(p); 1015 thread_stopped(p); 1016 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) { 1017 if (p->p_numthreads == p->p_suspcount + 1) { 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 if (wakeup_swapper) 1023 kick_proc0(); 1024 } 1025 } 1026 PROC_UNLOCK(p); 1027 thread_lock(td); 1028 /* 1029 * When a thread suspends, it just 1030 * gets taken off all queues. 1031 */ 1032 thread_suspend_one(td); 1033 if (return_instead == 0) { 1034 p->p_boundary_count++; 1035 td->td_flags |= TDF_BOUNDARY; 1036 } 1037 PROC_SUNLOCK(p); 1038 mi_switch(SW_INVOL | SWT_SUSPEND, NULL); 1039 thread_unlock(td); 1040 PROC_LOCK(p); 1041 } 1042 return (0); 1043} 1044 1045void 1046thread_suspend_switch(struct thread *td, struct proc *p) 1047{ 1048 1049 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 1050 PROC_LOCK_ASSERT(p, MA_OWNED); 1051 PROC_SLOCK_ASSERT(p, MA_OWNED); 1052 /* 1053 * We implement thread_suspend_one in stages here to avoid 1054 * dropping the proc lock while the thread lock is owned. 1055 */ 1056 if (p == td->td_proc) { 1057 thread_stopped(p); 1058 p->p_suspcount++; 1059 } 1060 PROC_UNLOCK(p); 1061 thread_lock(td); 1062 td->td_flags &= ~TDF_NEEDSUSPCHK; 1063 TD_SET_SUSPENDED(td); 1064 sched_sleep(td, 0); 1065 PROC_SUNLOCK(p); 1066 DROP_GIANT(); 1067 mi_switch(SW_VOL | SWT_SUSPEND, NULL); 1068 thread_unlock(td); 1069 PICKUP_GIANT(); 1070 PROC_LOCK(p); 1071 PROC_SLOCK(p); 1072} 1073 1074void 1075thread_suspend_one(struct thread *td) 1076{ 1077 struct proc *p; 1078 1079 p = td->td_proc; 1080 PROC_SLOCK_ASSERT(p, MA_OWNED); 1081 THREAD_LOCK_ASSERT(td, MA_OWNED); 1082 KASSERT(!TD_IS_SUSPENDED(td), ("already suspended")); 1083 p->p_suspcount++; 1084 td->td_flags &= ~TDF_NEEDSUSPCHK; 1085 TD_SET_SUSPENDED(td); 1086 sched_sleep(td, 0); 1087} 1088 1089static int 1090thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary) 1091{ 1092 1093 THREAD_LOCK_ASSERT(td, MA_OWNED); 1094 KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended")); 1095 TD_CLR_SUSPENDED(td); 1096 td->td_flags &= ~TDF_ALLPROCSUSP; 1097 if (td->td_proc == p) { 1098 PROC_SLOCK_ASSERT(p, MA_OWNED); 1099 p->p_suspcount--; 1100 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) { 1101 td->td_flags &= ~TDF_BOUNDARY; 1102 p->p_boundary_count--; 1103 } 1104 } 1105 return (setrunnable(td)); 1106} 1107 1108/* 1109 * Allow all threads blocked by single threading to continue running. 1110 */ 1111void 1112thread_unsuspend(struct proc *p) 1113{ 1114 struct thread *td; 1115 int wakeup_swapper; 1116 1117 PROC_LOCK_ASSERT(p, MA_OWNED); 1118 PROC_SLOCK_ASSERT(p, MA_OWNED); 1119 wakeup_swapper = 0; 1120 if (!P_SHOULDSTOP(p)) { 1121 FOREACH_THREAD_IN_PROC(p, td) { 1122 thread_lock(td); 1123 if (TD_IS_SUSPENDED(td)) { 1124 wakeup_swapper |= thread_unsuspend_one(td, p, 1125 true); 1126 } 1127 thread_unlock(td); 1128 } 1129 } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE && 1130 p->p_numthreads == p->p_suspcount) { 1131 /* 1132 * Stopping everything also did the job for the single 1133 * threading request. Now we've downgraded to single-threaded, 1134 * let it continue. 1135 */ 1136 if (p->p_singlethread->td_proc == p) { 1137 thread_lock(p->p_singlethread); 1138 wakeup_swapper = thread_unsuspend_one( 1139 p->p_singlethread, p, false); 1140 thread_unlock(p->p_singlethread); 1141 } 1142 } 1143 if (wakeup_swapper) 1144 kick_proc0(); 1145} 1146 1147/* 1148 * End the single threading mode.. 1149 */ 1150void 1151thread_single_end(struct proc *p, int mode) 1152{ 1153 struct thread *td; 1154 int wakeup_swapper; 1155 1156 KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY || 1157 mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT, 1158 ("invalid mode %d", mode)); 1159 PROC_LOCK_ASSERT(p, MA_OWNED); 1160 KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) || 1161 (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0), 1162 ("mode %d does not match P_TOTAL_STOP", mode)); 1163 KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread, 1164 ("thread_single_end from other thread %p %p", 1165 curthread, p->p_singlethread)); 1166 KASSERT(mode != SINGLE_BOUNDARY || 1167 (p->p_flag & P_SINGLE_BOUNDARY) != 0, 1168 ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag)); 1169 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY | 1170 P_TOTAL_STOP); 1171 PROC_SLOCK(p); 1172 p->p_singlethread = NULL; 1173 wakeup_swapper = 0; 1174 /* 1175 * If there are other threads they may now run, 1176 * unless of course there is a blanket 'stop order' 1177 * on the process. The single threader must be allowed 1178 * to continue however as this is a bad place to stop. 1179 */ 1180 if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) { 1181 FOREACH_THREAD_IN_PROC(p, td) { 1182 thread_lock(td); 1183 if (TD_IS_SUSPENDED(td)) { 1184 wakeup_swapper |= thread_unsuspend_one(td, p, 1185 mode == SINGLE_BOUNDARY); 1186 } 1187 thread_unlock(td); 1188 } 1189 } 1190 KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0, 1191 ("inconsistent boundary count %d", p->p_boundary_count)); 1192 PROC_SUNLOCK(p); 1193 if (wakeup_swapper) 1194 kick_proc0(); 1195} 1196 1197struct thread * 1198thread_find(struct proc *p, lwpid_t tid) 1199{ 1200 struct thread *td; 1201 1202 PROC_LOCK_ASSERT(p, MA_OWNED); 1203 FOREACH_THREAD_IN_PROC(p, td) { 1204 if (td->td_tid == tid) 1205 break; 1206 } 1207 return (td); 1208} 1209 1210/* Locate a thread by number; return with proc lock held. */ 1211struct thread * 1212tdfind(lwpid_t tid, pid_t pid) 1213{ 1214#define RUN_THRESH 16 1215 struct thread *td; 1216 int run = 0; 1217 1218 rw_rlock(&tidhash_lock); 1219 LIST_FOREACH(td, TIDHASH(tid), td_hash) { 1220 if (td->td_tid == tid) { 1221 if (pid != -1 && td->td_proc->p_pid != pid) { 1222 td = NULL; 1223 break; 1224 } 1225 PROC_LOCK(td->td_proc); 1226 if (td->td_proc->p_state == PRS_NEW) { 1227 PROC_UNLOCK(td->td_proc); 1228 td = NULL; 1229 break; 1230 } 1231 if (run > RUN_THRESH) { 1232 if (rw_try_upgrade(&tidhash_lock)) { 1233 LIST_REMOVE(td, td_hash); 1234 LIST_INSERT_HEAD(TIDHASH(td->td_tid), 1235 td, td_hash); 1236 rw_wunlock(&tidhash_lock); 1237 return (td); 1238 } 1239 } 1240 break; 1241 } 1242 run++; 1243 } 1244 rw_runlock(&tidhash_lock); 1245 return (td); 1246} 1247 1248void 1249tidhash_add(struct thread *td) 1250{ 1251 rw_wlock(&tidhash_lock); 1252 LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash); 1253 rw_wunlock(&tidhash_lock); 1254} 1255 1256void 1257tidhash_remove(struct thread *td) 1258{ 1259 rw_wlock(&tidhash_lock); 1260 LIST_REMOVE(td, td_hash); 1261 rw_wunlock(&tidhash_lock); 1262} 1263