sys/kern/kern_thread.c

   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/msan.h>
  42 #include <sys/mutex.h>
  43 #include <sys/proc.h>
  44 #include <sys/bitstring.h>
  45 #include <sys/epoch.h>
  46 #include <sys/rangelock.h>
  47 #include <sys/resourcevar.h>
  48 #include <sys/sdt.h>
  49 #include <sys/smp.h>
  50 #include <sys/sched.h>
  51 #include <sys/sleepqueue.h>
  52 #include <sys/selinfo.h>
  53 #include <sys/syscallsubr.h>
  54 #include <sys/dtrace_bsd.h>
  55 #include <sys/sysent.h>
  56 #include <sys/turnstile.h>
  57 #include <sys/taskqueue.h>
  58 #include <sys/ktr.h>
  59 #include <sys/rwlock.h>
  60 #include <sys/umtxvar.h>
  61 #include <sys/vmmeter.h>
  62 #include <sys/cpuset.h>
  63 #ifdef  HWPMC_HOOKS
  64 #include <sys/pmckern.h>
  65 #endif
  66 #include <sys/priv.h>
  67
  68 #include <security/audit/audit.h>
  69
  70 #include <vm/pmap.h>
  71 #include <vm/vm.h>
  72 #include <vm/vm_extern.h>
  73 #include <vm/uma.h>
  74 #include <vm/vm_phys.h>
  75 #include <sys/eventhandler.h>
  76
  77 /*
  78  * Asserts below verify the stability of struct thread and struct proc
  79  * layout, as exposed by KBI to modules.  On head, the KBI is allowed
  80  * to drift, change to the structures must be accompanied by the
  81  * assert update.
  82  *
  83  * On the stable branches after KBI freeze, conditions must not be
  84  * violated.  Typically new fields are moved to the end of the
  85  * structures.
  86  */
  87 #ifdef __amd64__
  88 _Static_assert(offsetof(struct thread, td_flags) == 0x108,
  89     "struct thread KBI td_flags");
  90 _Static_assert(offsetof(struct thread, td_pflags) == 0x110,
  91     "struct thread KBI td_pflags");
  92 _Static_assert(offsetof(struct thread, td_frame) == 0x4a8,
  93     "struct thread KBI td_frame");
  94 _Static_assert(offsetof(struct thread, td_emuldata) == 0x6b0,
  95     "struct thread KBI td_emuldata");
  96 _Static_assert(offsetof(struct proc, p_flag) == 0xb8,
  97     "struct proc KBI p_flag");
  98 _Static_assert(offsetof(struct proc, p_pid) == 0xc4,
  99     "struct proc KBI p_pid");
 100 _Static_assert(offsetof(struct proc, p_filemon) == 0x3c8,
 101     "struct proc KBI p_filemon");
 102 _Static_assert(offsetof(struct proc, p_comm) == 0x3e0,
 103     "struct proc KBI p_comm");
 104 _Static_assert(offsetof(struct proc, p_emuldata) == 0x4c8,
 105     "struct proc KBI p_emuldata");
 106 #endif
 107 #ifdef __i386__
 108 _Static_assert(offsetof(struct thread, td_flags) == 0x9c,
 109     "struct thread KBI td_flags");
 110 _Static_assert(offsetof(struct thread, td_pflags) == 0xa4,
 111     "struct thread KBI td_pflags");
 112 _Static_assert(offsetof(struct thread, td_frame) == 0x308,
 113     "struct thread KBI td_frame");
 114 _Static_assert(offsetof(struct thread, td_emuldata) == 0x34c,
 115     "struct thread KBI td_emuldata");
 116 _Static_assert(offsetof(struct proc, p_flag) == 0x6c,
 117     "struct proc KBI p_flag");
 118 _Static_assert(offsetof(struct proc, p_pid) == 0x78,
 119     "struct proc KBI p_pid");
 120 _Static_assert(offsetof(struct proc, p_filemon) == 0x270,
 121     "struct proc KBI p_filemon");
 122 _Static_assert(offsetof(struct proc, p_comm) == 0x284,
 123     "struct proc KBI p_comm");
 124 _Static_assert(offsetof(struct proc, p_emuldata) == 0x310,
 125     "struct proc KBI p_emuldata");
 126 #endif
 127
 128 SDT_PROVIDER_DECLARE(proc);
 129 SDT_PROBE_DEFINE(proc, , , lwp__exit);
 130
 131 /*
 132  * thread related storage.
 133  */
 134 static uma_zone_t thread_zone;
 135
 136 struct thread_domain_data {
 137         struct thread   *tdd_zombies;
 138         int             tdd_reapticks;
 139 } __aligned(CACHE_LINE_SIZE);
 140
 141 static struct thread_domain_data thread_domain_data[MAXMEMDOM];
 142
 143 static struct task      thread_reap_task;
 144 static struct callout   thread_reap_callout;
 145
 146 static void thread_zombie(struct thread *);
 147 static void thread_reap(void);
 148 static void thread_reap_all(void);
 149 static void thread_reap_task_cb(void *, int);
 150 static void thread_reap_callout_cb(void *);
 151 static int thread_unsuspend_one(struct thread *td, struct proc *p,
 152     bool boundary);
 153 static void thread_free_batched(struct thread *td);
 154
 155 static __exclusive_cache_line struct mtx tid_lock;
 156 static bitstr_t *tid_bitmap;
 157
 158 static MALLOC_DEFINE(M_TIDHASH, "tidhash", "thread hash");
 159
 160 static int maxthread;
 161 SYSCTL_INT(_kern, OID_AUTO, maxthread, CTLFLAG_RDTUN,
 162     &maxthread, 0, "Maximum number of threads");
 163
 164 static __exclusive_cache_line int nthreads;
 165
 166 static LIST_HEAD(tidhashhead, thread) *tidhashtbl;
 167 static u_long   tidhash;
 168 static u_long   tidhashlock;
 169 static struct   rwlock *tidhashtbl_lock;
 170 #define TIDHASH(tid)            (&tidhashtbl[(tid) & tidhash])
 171 #define TIDHASHLOCK(tid)        (&tidhashtbl_lock[(tid) & tidhashlock])
 172
 173 EVENTHANDLER_LIST_DEFINE(thread_ctor);
 174 EVENTHANDLER_LIST_DEFINE(thread_dtor);
 175 EVENTHANDLER_LIST_DEFINE(thread_init);
 176 EVENTHANDLER_LIST_DEFINE(thread_fini);
 177
 178 static bool
 179 thread_count_inc_try(void)
 180 {
 181         int nthreads_new;
 182
 183         nthreads_new = atomic_fetchadd_int(&nthreads, 1) + 1;
 184         if (nthreads_new >= maxthread - 100) {
 185                 if (priv_check_cred(curthread->td_ucred, PRIV_MAXPROC) != 0 ||
 186                     nthreads_new >= maxthread) {
 187                         atomic_subtract_int(&nthreads, 1);
 188                         return (false);
 189                 }
 190         }
 191         return (true);
 192 }
 193
 194 static bool
 195 thread_count_inc(void)
 196 {
 197         static struct timeval lastfail;
 198         static int curfail;
 199
 200         thread_reap();
 201         if (thread_count_inc_try()) {
 202                 return (true);
 203         }
 204
 205         thread_reap_all();
 206         if (thread_count_inc_try()) {
 207                 return (true);
 208         }
 209
 210         if (ppsratecheck(&lastfail, &curfail, 1)) {
 211                 printf("maxthread limit exceeded by uid %u "
 212                     "(pid %d); consider increasing kern.maxthread\n",
 213                     curthread->td_ucred->cr_ruid, curproc->p_pid);
 214         }
 215         return (false);
 216 }
 217
 218 static void
 219 thread_count_sub(int n)
 220 {
 221
 222         atomic_subtract_int(&nthreads, n);
 223 }
 224
 225 static void
 226 thread_count_dec(void)
 227 {
 228
 229         thread_count_sub(1);
 230 }
 231
 232 static lwpid_t
 233 tid_alloc(void)
 234 {
 235         static lwpid_t trytid;
 236         lwpid_t tid;
 237
 238         mtx_lock(&tid_lock);
 239         /*
 240          * It is an invariant that the bitmap is big enough to hold maxthread
 241          * IDs. If we got to this point there has to be at least one free.
 242          */
 243         if (trytid >= maxthread)
 244                 trytid = 0;
 245         bit_ffc_at(tid_bitmap, trytid, maxthread, &tid);
 246         if (tid == -1) {
 247                 KASSERT(trytid != 0, ("unexpectedly ran out of IDs"));
 248                 trytid = 0;
 249                 bit_ffc_at(tid_bitmap, trytid, maxthread, &tid);
 250                 KASSERT(tid != -1, ("unexpectedly ran out of IDs"));
 251         }
 252         bit_set(tid_bitmap, tid);
 253         trytid = tid + 1;
 254         mtx_unlock(&tid_lock);
 255         return (tid + NO_PID);
 256 }
 257
 258 static void
 259 tid_free_locked(lwpid_t rtid)
 260 {
 261         lwpid_t tid;
 262
 263         mtx_assert(&tid_lock, MA_OWNED);
 264         KASSERT(rtid >= NO_PID,
 265             ("%s: invalid tid %d\n", __func__, rtid));
 266         tid = rtid - NO_PID;
 267         KASSERT(bit_test(tid_bitmap, tid) != 0,
 268             ("thread ID %d not allocated\n", rtid));
 269         bit_clear(tid_bitmap, tid);
 270 }
 271
 272 static void
 273 tid_free(lwpid_t rtid)
 274 {
 275
 276         mtx_lock(&tid_lock);
 277         tid_free_locked(rtid);
 278         mtx_unlock(&tid_lock);
 279 }
 280
 281 static void
 282 tid_free_batch(lwpid_t *batch, int n)
 283 {
 284         int i;
 285
 286         mtx_lock(&tid_lock);
 287         for (i = 0; i < n; i++) {
 288                 tid_free_locked(batch[i]);
 289         }
 290         mtx_unlock(&tid_lock);
 291 }
 292
 293 /*
 294  * Batching for thread reapping.
 295  */
 296 struct tidbatch {
 297         lwpid_t tab[16];
 298         int n;
 299 };
 300
 301 static void
 302 tidbatch_prep(struct tidbatch *tb)
 303 {
 304
 305         tb->n = 0;
 306 }
 307
 308 static void
 309 tidbatch_add(struct tidbatch *tb, struct thread *td)
 310 {
 311
 312         KASSERT(tb->n < nitems(tb->tab),
 313             ("%s: count too high %d", __func__, tb->n));
 314         tb->tab[tb->n] = td->td_tid;
 315         tb->n++;
 316 }
 317
 318 static void
 319 tidbatch_process(struct tidbatch *tb)
 320 {
 321
 322         KASSERT(tb->n <= nitems(tb->tab),
 323             ("%s: count too high %d", __func__, tb->n));
 324         if (tb->n == nitems(tb->tab)) {
 325                 tid_free_batch(tb->tab, tb->n);
 326                 tb->n = 0;
 327         }
 328 }
 329
 330 static void
 331 tidbatch_final(struct tidbatch *tb)
 332 {
 333
 334         KASSERT(tb->n <= nitems(tb->tab),
 335             ("%s: count too high %d", __func__, tb->n));
 336         if (tb->n != 0) {
 337                 tid_free_batch(tb->tab, tb->n);
 338         }
 339 }
 340
 341 /*
 342  * Prepare a thread for use.
 343  */
 344 static int
 345 thread_ctor(void *mem, int size, void *arg, int flags)
 346 {
 347         struct thread   *td;
 348
 349         td = (struct thread *)mem;
 350         TD_SET_STATE(td, TDS_INACTIVE);
 351         td->td_lastcpu = td->td_oncpu = NOCPU;
 352
 353         /*
 354          * Note that td_critnest begins life as 1 because the thread is not
 355          * running and is thereby implicitly waiting to be on the receiving
 356          * end of a context switch.
 357          */
 358         td->td_critnest = 1;
 359         td->td_lend_user_pri = PRI_MAX;
 360 #ifdef AUDIT
 361         audit_thread_alloc(td);
 362 #endif
 363 #ifdef KDTRACE_HOOKS
 364         kdtrace_thread_ctor(td);
 365 #endif
 366         umtx_thread_alloc(td);
 367         MPASS(td->td_sel == NULL);
 368         return (0);
 369 }
 370
 371 /*
 372  * Reclaim a thread after use.
 373  */
 374 static void
 375 thread_dtor(void *mem, int size, void *arg)
 376 {
 377         struct thread *td;
 378
 379         td = (struct thread *)mem;
 380
 381 #ifdef INVARIANTS
 382         /* Verify that this thread is in a safe state to free. */
 383         switch (TD_GET_STATE(td)) {
 384         case TDS_INHIBITED:
 385         case TDS_RUNNING:
 386         case TDS_CAN_RUN:
 387         case TDS_RUNQ:
 388                 /*
 389                  * We must never unlink a thread that is in one of
 390                  * these states, because it is currently active.
 391                  */
 392                 panic("bad state for thread unlinking");
 393                 /* NOTREACHED */
 394         case TDS_INACTIVE:
 395                 break;
 396         default:
 397                 panic("bad thread state");
 398                 /* NOTREACHED */
 399         }
 400 #endif
 401 #ifdef AUDIT
 402         audit_thread_free(td);
 403 #endif
 404 #ifdef KDTRACE_HOOKS
 405         kdtrace_thread_dtor(td);
 406 #endif
 407         /* Free all OSD associated to this thread. */
 408         osd_thread_exit(td);
 409         td_softdep_cleanup(td);
 410         MPASS(td->td_su == NULL);
 411         seltdfini(td);
 412 }
 413
 414 /*
 415  * Initialize type-stable parts of a thread (when newly created).
 416  */
 417 static int
 418 thread_init(void *mem, int size, int flags)
 419 {
 420         struct thread *td;
 421
 422         td = (struct thread *)mem;
 423
 424         td->td_allocdomain = vm_phys_domain(vtophys(td));
 425         td->td_sleepqueue = sleepq_alloc();
 426         td->td_turnstile = turnstile_alloc();
 427         td->td_rlqe = NULL;
 428         EVENTHANDLER_DIRECT_INVOKE(thread_init, td);
 429         umtx_thread_init(td);
 430         td->td_kstack = 0;
 431         td->td_sel = NULL;
 432         return (0);
 433 }
 434
 435 /*
 436  * Tear down type-stable parts of a thread (just before being discarded).
 437  */
 438 static void
 439 thread_fini(void *mem, int size)
 440 {
 441         struct thread *td;
 442
 443         td = (struct thread *)mem;
 444         EVENTHANDLER_DIRECT_INVOKE(thread_fini, td);
 445         rlqentry_free(td->td_rlqe);
 446         turnstile_free(td->td_turnstile);
 447         sleepq_free(td->td_sleepqueue);
 448         umtx_thread_fini(td);
 449         MPASS(td->td_sel == NULL);
 450 }
 451
 452 /*
 453  * For a newly created process,
 454  * link up all the structures and its initial threads etc.
 455  * called from:
 456  * {arch}/{arch}/machdep.c   {arch}_init(), init386() etc.
 457  * proc_dtor() (should go away)
 458  * proc_init()
 459  */
 460 void
 461 proc_linkup0(struct proc *p, struct thread *td)
 462 {
 463         TAILQ_INIT(&p->p_threads);           /* all threads in proc */
 464         proc_linkup(p, td);
 465 }
 466
 467 void
 468 proc_linkup(struct proc *p, struct thread *td)
 469 {
 470
 471         sigqueue_init(&p->p_sigqueue, p);
 472         p->p_ksi = ksiginfo_alloc(1);
 473         if (p->p_ksi != NULL) {
 474                 /* XXX p_ksi may be null if ksiginfo zone is not ready */
 475                 p->p_ksi->ksi_flags = KSI_EXT | KSI_INS;
 476         }
 477         LIST_INIT(&p->p_mqnotifier);
 478         p->p_numthreads = 0;
 479         thread_link(td, p);
 480 }
 481
 482 extern int max_threads_per_proc;
 483
 484 /*
 485  * Initialize global thread allocation resources.
 486  */
 487 void
 488 threadinit(void)
 489 {
 490         u_long i;
 491         lwpid_t tid0;
 492         uint32_t flags;
 493
 494         /*
 495          * Place an upper limit on threads which can be allocated.
 496          *
 497          * Note that other factors may make the de facto limit much lower.
 498          *
 499          * Platform limits are somewhat arbitrary but deemed "more than good
 500          * enough" for the foreseable future.
 501          */
 502         if (maxthread == 0) {
 503 #ifdef _LP64
 504                 maxthread = MIN(maxproc * max_threads_per_proc, 1000000);
 505 #else
 506                 maxthread = MIN(maxproc * max_threads_per_proc, 100000);
 507 #endif
 508         }
 509
 510         mtx_init(&tid_lock, "TID lock", NULL, MTX_DEF);
 511         tid_bitmap = bit_alloc(maxthread, M_TIDHASH, M_WAITOK);
 512         /*
 513          * Handle thread0.
 514          */
 515         thread_count_inc();
 516         tid0 = tid_alloc();
 517         if (tid0 != THREAD0_TID)
 518                 panic("tid0 %d != %d\n", tid0, THREAD0_TID);
 519
 520         flags = UMA_ZONE_NOFREE;
 521 #ifdef __aarch64__
 522         /*
 523          * Force thread structures to be allocated from the direct map.
 524          * Otherwise, superpage promotions and demotions may temporarily
 525          * invalidate thread structure mappings.  For most dynamically allocated
 526          * structures this is not a problem, but translation faults cannot be
 527          * handled without accessing curthread.
 528          */
 529         flags |= UMA_ZONE_CONTIG;
 530 #endif
 531         thread_zone = uma_zcreate("THREAD", sched_sizeof_thread(),
 532             thread_ctor, thread_dtor, thread_init, thread_fini,
 533             32 - 1, flags);
 534         tidhashtbl = hashinit(maxproc / 2, M_TIDHASH, &tidhash);
 535         tidhashlock = (tidhash + 1) / 64;
 536         if (tidhashlock > 0)
 537                 tidhashlock--;
 538         tidhashtbl_lock = malloc(sizeof(*tidhashtbl_lock) * (tidhashlock + 1),
 539             M_TIDHASH, M_WAITOK | M_ZERO);
 540         for (i = 0; i < tidhashlock + 1; i++)
 541                 rw_init(&tidhashtbl_lock[i], "tidhash");
 542
 543         TASK_INIT(&thread_reap_task, 0, thread_reap_task_cb, NULL);
 544         callout_init(&thread_reap_callout, 1);
 545         callout_reset(&thread_reap_callout, 5 * hz,
 546             thread_reap_callout_cb, NULL);
 547 }
 548
 549 /*
 550  * Place an unused thread on the zombie list.
 551  */
 552 void
 553 thread_zombie(struct thread *td)
 554 {
 555         struct thread_domain_data *tdd;
 556         struct thread *ztd;
 557
 558         tdd = &thread_domain_data[td->td_allocdomain];
 559         ztd = atomic_load_ptr(&tdd->tdd_zombies);
 560         for (;;) {
 561                 td->td_zombie = ztd;
 562                 if (atomic_fcmpset_rel_ptr((uintptr_t *)&tdd->tdd_zombies,
 563                     (uintptr_t *)&ztd, (uintptr_t)td))
 564                         break;
 565                 continue;
 566         }
 567 }
 568
 569 /*
 570  * Release a thread that has exited after cpu_throw().
 571  */
 572 void
 573 thread_stash(struct thread *td)
 574 {
 575         atomic_subtract_rel_int(&td->td_proc->p_exitthreads, 1);
 576         thread_zombie(td);
 577 }
 578
 579 /*
 580  * Reap zombies from passed domain.
 581  */
 582 static void
 583 thread_reap_domain(struct thread_domain_data *tdd)
 584 {
 585         struct thread *itd, *ntd;
 586         struct tidbatch tidbatch;
 587         struct credbatch credbatch;
 588         int tdcount;
 589         struct plimit *lim;
 590         int limcount;
 591
 592         /*
 593          * Reading upfront is pessimal if followed by concurrent atomic_swap,
 594          * but most of the time the list is empty.
 595          */
 596         if (tdd->tdd_zombies == NULL)
 597                 return;
 598
 599         itd = (struct thread *)atomic_swap_ptr((uintptr_t *)&tdd->tdd_zombies,
 600             (uintptr_t)NULL);
 601         if (itd == NULL)
 602                 return;
 603
 604         /*
 605          * Multiple CPUs can get here, the race is fine as ticks is only
 606          * advisory.
 607          */
 608         tdd->tdd_reapticks = ticks;
 609
 610         tidbatch_prep(&tidbatch);
 611         credbatch_prep(&credbatch);
 612         tdcount = 0;
 613         lim = NULL;
 614         limcount = 0;
 615
 616         while (itd != NULL) {
 617                 ntd = itd->td_zombie;
 618                 EVENTHANDLER_DIRECT_INVOKE(thread_dtor, itd);
 619                 tidbatch_add(&tidbatch, itd);
 620                 credbatch_add(&credbatch, itd);
 621                 MPASS(itd->td_limit != NULL);
 622                 if (lim != itd->td_limit) {
 623                         if (limcount != 0) {
 624                                 lim_freen(lim, limcount);
 625                                 limcount = 0;
 626                         }
 627                 }
 628                 lim = itd->td_limit;
 629                 limcount++;
 630                 thread_free_batched(itd);
 631                 tidbatch_process(&tidbatch);
 632                 credbatch_process(&credbatch);
 633                 tdcount++;
 634                 if (tdcount == 32) {
 635                         thread_count_sub(tdcount);
 636                         tdcount = 0;
 637                 }
 638                 itd = ntd;
 639         }
 640
 641         tidbatch_final(&tidbatch);
 642         credbatch_final(&credbatch);
 643         if (tdcount != 0) {
 644                 thread_count_sub(tdcount);
 645         }
 646         MPASS(limcount != 0);
 647         lim_freen(lim, limcount);
 648 }
 649
 650 /*
 651  * Reap zombies from all domains.
 652  */
 653 static void
 654 thread_reap_all(void)
 655 {
 656         struct thread_domain_data *tdd;
 657         int i, domain;
 658
 659         domain = PCPU_GET(domain);
 660         for (i = 0; i < vm_ndomains; i++) {
 661                 tdd = &thread_domain_data[(i + domain) % vm_ndomains];
 662                 thread_reap_domain(tdd);
 663         }
 664 }
 665
 666 /*
 667  * Reap zombies from local domain.
 668  */
 669 static void
 670 thread_reap(void)
 671 {
 672         struct thread_domain_data *tdd;
 673         int domain;
 674
 675         domain = PCPU_GET(domain);
 676         tdd = &thread_domain_data[domain];
 677
 678         thread_reap_domain(tdd);
 679 }
 680
 681 static void
 682 thread_reap_task_cb(void *arg __unused, int pending __unused)
 683 {
 684
 685         thread_reap_all();
 686 }
 687
 688 static void
 689 thread_reap_callout_cb(void *arg __unused)
 690 {
 691         struct thread_domain_data *tdd;
 692         int i, cticks, lticks;
 693         bool wantreap;
 694
 695         wantreap = false;
 696         cticks = atomic_load_int(&ticks);
 697         for (i = 0; i < vm_ndomains; i++) {
 698                 tdd = &thread_domain_data[i];
 699                 lticks = tdd->tdd_reapticks;
 700                 if (tdd->tdd_zombies != NULL &&
 701                     (u_int)(cticks - lticks) > 5 * hz) {
 702                         wantreap = true;
 703                         break;
 704                 }
 705         }
 706
 707         if (wantreap)
 708                 taskqueue_enqueue(taskqueue_thread, &thread_reap_task);
 709         callout_reset(&thread_reap_callout, 5 * hz,
 710             thread_reap_callout_cb, NULL);
 711 }
 712
 713 /*
 714  * Calling this function guarantees that any thread that exited before
 715  * the call is reaped when the function returns.  By 'exited' we mean
 716  * a thread removed from the process linkage with thread_unlink().
 717  * Practically this means that caller must lock/unlock corresponding
 718  * process lock before the call, to synchronize with thread_exit().
 719  */
 720 void
 721 thread_reap_barrier(void)
 722 {
 723         struct task *t;
 724
 725         /*
 726          * First do context switches to each CPU to ensure that all
 727          * PCPU pc_deadthreads are moved to zombie list.
 728          */
 729         quiesce_all_cpus("", PDROP);
 730
 731         /*
 732          * Second, fire the task in the same thread as normal
 733          * thread_reap() is done, to serialize reaping.
 734          */
 735         t = malloc(sizeof(*t), M_TEMP, M_WAITOK);
 736         TASK_INIT(t, 0, thread_reap_task_cb, t);
 737         taskqueue_enqueue(taskqueue_thread, t);
 738         taskqueue_drain(taskqueue_thread, t);
 739         free(t, M_TEMP);
 740 }
 741
 742 /*
 743  * Allocate a thread.
 744  */
 745 struct thread *
 746 thread_alloc(int pages)
 747 {
 748         struct thread *td;
 749         lwpid_t tid;
 750
 751         if (!thread_count_inc()) {
 752                 return (NULL);
 753         }
 754
 755         tid = tid_alloc();
 756         td = uma_zalloc(thread_zone, M_WAITOK);
 757         KASSERT(td->td_kstack == 0, ("thread_alloc got thread with kstack"));
 758         if (!vm_thread_new(td, pages)) {
 759                 uma_zfree(thread_zone, td);
 760                 tid_free(tid);
 761                 thread_count_dec();
 762                 return (NULL);
 763         }
 764         td->td_tid = tid;
 765         bzero(&td->td_sa.args, sizeof(td->td_sa.args));
 766         kmsan_thread_alloc(td);
 767         cpu_thread_alloc(td);
 768         EVENTHANDLER_DIRECT_INVOKE(thread_ctor, td);
 769         return (td);
 770 }
 771
 772 int
 773 thread_alloc_stack(struct thread *td, int pages)
 774 {
 775
 776         KASSERT(td->td_kstack == 0,
 777             ("thread_alloc_stack called on a thread with kstack"));
 778         if (!vm_thread_new(td, pages))
 779                 return (0);
 780         cpu_thread_alloc(td);
 781         return (1);
 782 }
 783
 784 /*
 785  * Deallocate a thread.
 786  */
 787 static void
 788 thread_free_batched(struct thread *td)
 789 {
 790
 791         lock_profile_thread_exit(td);
 792         if (td->td_cpuset)
 793                 cpuset_rel(td->td_cpuset);
 794         td->td_cpuset = NULL;
 795         cpu_thread_free(td);
 796         if (td->td_kstack != 0)
 797                 vm_thread_dispose(td);
 798         callout_drain(&td->td_slpcallout);
 799         /*
 800          * Freeing handled by the caller.
 801          */
 802         td->td_tid = -1;
 803         kmsan_thread_free(td);
 804         uma_zfree(thread_zone, td);
 805 }
 806
 807 void
 808 thread_free(struct thread *td)
 809 {
 810         lwpid_t tid;
 811
 812         EVENTHANDLER_DIRECT_INVOKE(thread_dtor, td);
 813         tid = td->td_tid;
 814         thread_free_batched(td);
 815         tid_free(tid);
 816         thread_count_dec();
 817 }
 818
 819 void
 820 thread_cow_get_proc(struct thread *newtd, struct proc *p)
 821 {
 822
 823         PROC_LOCK_ASSERT(p, MA_OWNED);
 824         newtd->td_realucred = crcowget(p->p_ucred);
 825         newtd->td_ucred = newtd->td_realucred;
 826         newtd->td_limit = lim_hold(p->p_limit);
 827         newtd->td_cowgen = p->p_cowgen;
 828 }
 829
 830 void
 831 thread_cow_get(struct thread *newtd, struct thread *td)
 832 {
 833
 834         MPASS(td->td_realucred == td->td_ucred);
 835         newtd->td_realucred = crcowget(td->td_realucred);
 836         newtd->td_ucred = newtd->td_realucred;
 837         newtd->td_limit = lim_hold(td->td_limit);
 838         newtd->td_cowgen = td->td_cowgen;
 839 }
 840
 841 void
 842 thread_cow_free(struct thread *td)
 843 {
 844
 845         if (td->td_realucred != NULL)
 846                 crcowfree(td);
 847         if (td->td_limit != NULL)
 848                 lim_free(td->td_limit);
 849 }
 850
 851 void
 852 thread_cow_update(struct thread *td)
 853 {
 854         struct proc *p;
 855         struct ucred *oldcred;
 856         struct plimit *oldlimit;
 857
 858         p = td->td_proc;
 859         PROC_LOCK(p);
 860         oldcred = crcowsync();
 861         oldlimit = lim_cowsync();
 862         td->td_cowgen = p->p_cowgen;
 863         PROC_UNLOCK(p);
 864         if (oldcred != NULL)
 865                 crfree(oldcred);
 866         if (oldlimit != NULL)
 867                 lim_free(oldlimit);
 868 }
 869
 870 void
 871 thread_cow_synced(struct thread *td)
 872 {
 873         struct proc *p;
 874
 875         p = td->td_proc;
 876         PROC_LOCK_ASSERT(p, MA_OWNED);
 877         MPASS(td->td_cowgen != p->p_cowgen);
 878         MPASS(td->td_ucred == p->p_ucred);
 879         MPASS(td->td_limit == p->p_limit);
 880         td->td_cowgen = p->p_cowgen;
 881 }
 882
 883 /*
 884  * Discard the current thread and exit from its context.
 885  * Always called with scheduler locked.
 886  *
 887  * Because we can't free a thread while we're operating under its context,
 888  * push the current thread into our CPU's deadthread holder. This means
 889  * we needn't worry about someone else grabbing our context before we
 890  * do a cpu_throw().
 891  */
 892 void
 893 thread_exit(void)
 894 {
 895         uint64_t runtime, new_switchtime;
 896         struct thread *td;
 897         struct thread *td2;
 898         struct proc *p;
 899         int wakeup_swapper;
 900
 901         td = curthread;
 902         p = td->td_proc;
 903
 904         PROC_SLOCK_ASSERT(p, MA_OWNED);
 905         mtx_assert(&Giant, MA_NOTOWNED);
 906
 907         PROC_LOCK_ASSERT(p, MA_OWNED);
 908         KASSERT(p != NULL, ("thread exiting without a process"));
 909         CTR3(KTR_PROC, "thread_exit: thread %p (pid %ld, %s)", td,
 910             (long)p->p_pid, td->td_name);
 911         SDT_PROBE0(proc, , , lwp__exit);
 912         KASSERT(TAILQ_EMPTY(&td->td_sigqueue.sq_list), ("signal pending"));
 913         MPASS(td->td_realucred == td->td_ucred);
 914
 915         /*
 916          * drop FPU & debug register state storage, or any other
 917          * architecture specific resources that
 918          * would not be on a new untouched process.
 919          */
 920         cpu_thread_exit(td);
 921
 922         /*
 923          * The last thread is left attached to the process
 924          * So that the whole bundle gets recycled. Skip
 925          * all this stuff if we never had threads.
 926          * EXIT clears all sign of other threads when
 927          * it goes to single threading, so the last thread always
 928          * takes the short path.
 929          */
 930         if (p->p_flag & P_HADTHREADS) {
 931                 if (p->p_numthreads > 1) {
 932                         atomic_add_int(&td->td_proc->p_exitthreads, 1);
 933                         thread_unlink(td);
 934                         td2 = FIRST_THREAD_IN_PROC(p);
 935                         sched_exit_thread(td2, td);
 936
 937                         /*
 938                          * The test below is NOT true if we are the
 939                          * sole exiting thread. P_STOPPED_SINGLE is unset
 940                          * in exit1() after it is the only survivor.
 941                          */
 942                         if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
 943                                 if (p->p_numthreads == p->p_suspcount) {
 944                                         thread_lock(p->p_singlethread);
 945                                         wakeup_swapper = thread_unsuspend_one(
 946                                                 p->p_singlethread, p, false);
 947                                         if (wakeup_swapper)
 948                                                 kick_proc0();
 949                                 }
 950                         }
 951
 952                         PCPU_SET(deadthread, td);
 953                 } else {
 954                         /*
 955                          * The last thread is exiting.. but not through exit()
 956                          */
 957                         panic ("thread_exit: Last thread exiting on its own");
 958                 }
 959         }
 960 #ifdef  HWPMC_HOOKS
 961         /*
 962          * If this thread is part of a process that is being tracked by hwpmc(4),
 963          * inform the module of the thread's impending exit.
 964          */
 965         if (PMC_PROC_IS_USING_PMCS(td->td_proc)) {
 966                 PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT);
 967                 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT, NULL);
 968         } else if (PMC_SYSTEM_SAMPLING_ACTIVE())
 969                 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT_LOG, NULL);
 970 #endif
 971         PROC_UNLOCK(p);
 972         PROC_STATLOCK(p);
 973         thread_lock(td);
 974         PROC_SUNLOCK(p);
 975
 976         /* Do the same timestamp bookkeeping that mi_switch() would do. */
 977         new_switchtime = cpu_ticks();
 978         runtime = new_switchtime - PCPU_GET(switchtime);
 979         td->td_runtime += runtime;
 980         td->td_incruntime += runtime;
 981         PCPU_SET(switchtime, new_switchtime);
 982         PCPU_SET(switchticks, ticks);
 983         VM_CNT_INC(v_swtch);
 984
 985         /* Save our resource usage in our process. */
 986         td->td_ru.ru_nvcsw++;
 987         ruxagg_locked(p, td);
 988         rucollect(&p->p_ru, &td->td_ru);
 989         PROC_STATUNLOCK(p);
 990
 991         TD_SET_STATE(td, TDS_INACTIVE);
 992 #ifdef WITNESS
 993         witness_thread_exit(td);
 994 #endif
 995         CTR1(KTR_PROC, "thread_exit: cpu_throw() thread %p", td);
 996         sched_throw(td);
 997         panic("I'm a teapot!");
 998         /* NOTREACHED */
 999 }
1000
1001 /*
1002  * Do any thread specific cleanups that may be needed in wait()
1003  * called with Giant, proc and schedlock not held.
1004  */
1005 void
1006 thread_wait(struct proc *p)
1007 {
1008         struct thread *td;
1009
1010         mtx_assert(&Giant, MA_NOTOWNED);
1011         KASSERT(p->p_numthreads == 1, ("multiple threads in thread_wait()"));
1012         KASSERT(p->p_exitthreads == 0, ("p_exitthreads leaking"));
1013         td = FIRST_THREAD_IN_PROC(p);
1014         /* Lock the last thread so we spin until it exits cpu_throw(). */
1015         thread_lock(td);
1016         thread_unlock(td);
1017         lock_profile_thread_exit(td);
1018         cpuset_rel(td->td_cpuset);
1019         td->td_cpuset = NULL;
1020         cpu_thread_clean(td);
1021         thread_cow_free(td);
1022         callout_drain(&td->td_slpcallout);
1023         thread_reap();  /* check for zombie threads etc. */
1024 }
1025
1026 /*
1027  * Link a thread to a process.
1028  * set up anything that needs to be initialized for it to
1029  * be used by the process.
1030  */
1031 void
1032 thread_link(struct thread *td, struct proc *p)
1033 {
1034
1035         /*
1036          * XXX This can't be enabled because it's called for proc0 before
1037          * its lock has been created.
1038          * PROC_LOCK_ASSERT(p, MA_OWNED);
1039          */
1040         TD_SET_STATE(td, TDS_INACTIVE);
1041         td->td_proc     = p;
1042         td->td_flags    = TDF_INMEM;
1043
1044         LIST_INIT(&td->td_contested);
1045         LIST_INIT(&td->td_lprof[0]);
1046         LIST_INIT(&td->td_lprof[1]);
1047 #ifdef EPOCH_TRACE
1048         SLIST_INIT(&td->td_epochs);
1049 #endif
1050         sigqueue_init(&td->td_sigqueue, p);
1051         callout_init(&td->td_slpcallout, 1);
1052         TAILQ_INSERT_TAIL(&p->p_threads, td, td_plist);
1053         p->p_numthreads++;
1054 }
1055
1056 /*
1057  * Called from:
1058  *  thread_exit()
1059  */
1060 void
1061 thread_unlink(struct thread *td)
1062 {
1063         struct proc *p = td->td_proc;
1064
1065         PROC_LOCK_ASSERT(p, MA_OWNED);
1066 #ifdef EPOCH_TRACE
1067         MPASS(SLIST_EMPTY(&td->td_epochs));
1068 #endif
1069
1070         TAILQ_REMOVE(&p->p_threads, td, td_plist);
1071         p->p_numthreads--;
1072         /* could clear a few other things here */
1073         /* Must  NOT clear links to proc! */
1074 }
1075
1076 static int
1077 calc_remaining(struct proc *p, int mode)
1078 {
1079         int remaining;
1080
1081         PROC_LOCK_ASSERT(p, MA_OWNED);
1082         PROC_SLOCK_ASSERT(p, MA_OWNED);
1083         if (mode == SINGLE_EXIT)
1084                 remaining = p->p_numthreads;
1085         else if (mode == SINGLE_BOUNDARY)
1086                 remaining = p->p_numthreads - p->p_boundary_count;
1087         else if (mode == SINGLE_NO_EXIT || mode == SINGLE_ALLPROC)
1088                 remaining = p->p_numthreads - p->p_suspcount;
1089         else
1090                 panic("calc_remaining: wrong mode %d", mode);
1091         return (remaining);
1092 }
1093
1094 static int
1095 remain_for_mode(int mode)
1096 {
1097
1098         return (mode == SINGLE_ALLPROC ? 0 : 1);
1099 }
1100
1101 static int
1102 weed_inhib(int mode, struct thread *td2, struct proc *p)
1103 {
1104         int wakeup_swapper;
1105
1106         PROC_LOCK_ASSERT(p, MA_OWNED);
1107         PROC_SLOCK_ASSERT(p, MA_OWNED);
1108         THREAD_LOCK_ASSERT(td2, MA_OWNED);
1109
1110         wakeup_swapper = 0;
1111
1112         /*
1113          * Since the thread lock is dropped by the scheduler we have
1114          * to retry to check for races.
1115          */
1116 restart:
1117         switch (mode) {
1118         case SINGLE_EXIT:
1119                 if (TD_IS_SUSPENDED(td2)) {
1120                         wakeup_swapper |= thread_unsuspend_one(td2, p, true);
1121                         thread_lock(td2);
1122                         goto restart;
1123                 }
1124                 if (TD_CAN_ABORT(td2)) {
1125                         wakeup_swapper |= sleepq_abort(td2, EINTR);
1126                         return (wakeup_swapper);
1127                 }
1128                 break;
1129         case SINGLE_BOUNDARY:
1130         case SINGLE_NO_EXIT:
1131                 if (TD_IS_SUSPENDED(td2) &&
1132                     (td2->td_flags & TDF_BOUNDARY) == 0) {
1133                         wakeup_swapper |= thread_unsuspend_one(td2, p, false);
1134                         thread_lock(td2);
1135                         goto restart;
1136                 }
1137                 if (TD_CAN_ABORT(td2)) {
1138                         wakeup_swapper |= sleepq_abort(td2, ERESTART);
1139                         return (wakeup_swapper);
1140                 }
1141                 break;
1142         case SINGLE_ALLPROC:
1143                 /*
1144                  * ALLPROC suspend tries to avoid spurious EINTR for
1145                  * threads sleeping interruptable, by suspending the
1146                  * thread directly, similarly to sig_suspend_threads().
1147                  * Since such sleep is not performed at the user
1148                  * boundary, TDF_BOUNDARY flag is not set, and TDF_ALLPROCSUSP
1149                  * is used to avoid immediate un-suspend.
1150                  */
1151                 if (TD_IS_SUSPENDED(td2) && (td2->td_flags & (TDF_BOUNDARY |
1152                     TDF_ALLPROCSUSP)) == 0) {
1153                         wakeup_swapper |= thread_unsuspend_one(td2, p, false);
1154                         thread_lock(td2);
1155                         goto restart;
1156                 }
1157                 if (TD_CAN_ABORT(td2)) {
1158                         if ((td2->td_flags & TDF_SBDRY) == 0) {
1159                                 thread_suspend_one(td2);
1160                                 td2->td_flags |= TDF_ALLPROCSUSP;
1161                         } else {
1162                                 wakeup_swapper |= sleepq_abort(td2, ERESTART);
1163                                 return (wakeup_swapper);
1164                         }
1165                 }
1166                 break;
1167         default:
1168                 break;
1169         }
1170         thread_unlock(td2);
1171         return (wakeup_swapper);
1172 }
1173
1174 /*
1175  * Enforce single-threading.
1176  *
1177  * Returns 1 if the caller must abort (another thread is waiting to
1178  * exit the process or similar). Process is locked!
1179  * Returns 0 when you are successfully the only thread running.
1180  * A process has successfully single threaded in the suspend mode when
1181  * There are no threads in user mode. Threads in the kernel must be
1182  * allowed to continue until they get to the user boundary. They may even
1183  * copy out their return values and data before suspending. They may however be
1184  * accelerated in reaching the user boundary as we will wake up
1185  * any sleeping threads that are interruptable. (PCATCH).
1186  */
1187 int
1188 thread_single(struct proc *p, int mode)
1189 {
1190         struct thread *td;
1191         struct thread *td2;
1192         int remaining, wakeup_swapper;
1193
1194         td = curthread;
1195         KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
1196             mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
1197             ("invalid mode %d", mode));
1198         /*
1199          * If allowing non-ALLPROC singlethreading for non-curproc
1200          * callers, calc_remaining() and remain_for_mode() should be
1201          * adjusted to also account for td->td_proc != p.  For now
1202          * this is not implemented because it is not used.
1203          */
1204         KASSERT((mode == SINGLE_ALLPROC && td->td_proc != p) ||
1205             (mode != SINGLE_ALLPROC && td->td_proc == p),
1206             ("mode %d proc %p curproc %p", mode, p, td->td_proc));
1207         mtx_assert(&Giant, MA_NOTOWNED);
1208         PROC_LOCK_ASSERT(p, MA_OWNED);
1209
1210         if ((p->p_flag & P_HADTHREADS) == 0 && mode != SINGLE_ALLPROC)
1211                 return (0);
1212
1213         /* Is someone already single threading? */
1214         if (p->p_singlethread != NULL && p->p_singlethread != td)
1215                 return (1);
1216
1217         if (mode == SINGLE_EXIT) {
1218                 p->p_flag |= P_SINGLE_EXIT;
1219                 p->p_flag &= ~P_SINGLE_BOUNDARY;
1220         } else {
1221                 p->p_flag &= ~P_SINGLE_EXIT;
1222                 if (mode == SINGLE_BOUNDARY)
1223                         p->p_flag |= P_SINGLE_BOUNDARY;
1224                 else
1225                         p->p_flag &= ~P_SINGLE_BOUNDARY;
1226         }
1227         if (mode == SINGLE_ALLPROC)
1228                 p->p_flag |= P_TOTAL_STOP;
1229         p->p_flag |= P_STOPPED_SINGLE;
1230         PROC_SLOCK(p);
1231         p->p_singlethread = td;
1232         remaining = calc_remaining(p, mode);
1233         while (remaining != remain_for_mode(mode)) {
1234                 if (P_SHOULDSTOP(p) != P_STOPPED_SINGLE)
1235                         goto stopme;
1236                 wakeup_swapper = 0;
1237                 FOREACH_THREAD_IN_PROC(p, td2) {
1238                         if (td2 == td)
1239                                 continue;
1240                         thread_lock(td2);
1241                         td2->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
1242                         if (TD_IS_INHIBITED(td2)) {
1243                                 wakeup_swapper |= weed_inhib(mode, td2, p);
1244 #ifdef SMP
1245                         } else if (TD_IS_RUNNING(td2) && td != td2) {
1246                                 forward_signal(td2);
1247                                 thread_unlock(td2);
1248 #endif
1249                         } else
1250                                 thread_unlock(td2);
1251                 }
1252                 if (wakeup_swapper)
1253                         kick_proc0();
1254                 remaining = calc_remaining(p, mode);
1255
1256                 /*
1257                  * Maybe we suspended some threads.. was it enough?
1258                  */
1259                 if (remaining == remain_for_mode(mode))
1260                         break;
1261
1262 stopme:
1263                 /*
1264                  * Wake us up when everyone else has suspended.
1265                  * In the mean time we suspend as well.
1266                  */
1267                 thread_suspend_switch(td, p);
1268                 remaining = calc_remaining(p, mode);
1269         }
1270         if (mode == SINGLE_EXIT) {
1271                 /*
1272                  * Convert the process to an unthreaded process.  The
1273                  * SINGLE_EXIT is called by exit1() or execve(), in
1274                  * both cases other threads must be retired.
1275                  */
1276                 KASSERT(p->p_numthreads == 1, ("Unthreading with >1 threads"));
1277                 p->p_singlethread = NULL;
1278                 p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_HADTHREADS);
1279
1280                 /*
1281                  * Wait for any remaining threads to exit cpu_throw().
1282                  */
1283                 while (p->p_exitthreads != 0) {
1284                         PROC_SUNLOCK(p);
1285                         PROC_UNLOCK(p);
1286                         sched_relinquish(td);
1287                         PROC_LOCK(p);
1288                         PROC_SLOCK(p);
1289                 }
1290         } else if (mode == SINGLE_BOUNDARY) {
1291                 /*
1292                  * Wait until all suspended threads are removed from
1293                  * the processors.  The thread_suspend_check()
1294                  * increments p_boundary_count while it is still
1295                  * running, which makes it possible for the execve()
1296                  * to destroy vmspace while our other threads are
1297                  * still using the address space.
1298                  *
1299                  * We lock the thread, which is only allowed to
1300                  * succeed after context switch code finished using
1301                  * the address space.
1302                  */
1303                 FOREACH_THREAD_IN_PROC(p, td2) {
1304                         if (td2 == td)
1305                                 continue;
1306                         thread_lock(td2);
1307                         KASSERT((td2->td_flags & TDF_BOUNDARY) != 0,
1308                             ("td %p not on boundary", td2));
1309                         KASSERT(TD_IS_SUSPENDED(td2),
1310                             ("td %p is not suspended", td2));
1311                         thread_unlock(td2);
1312                 }
1313         }
1314         PROC_SUNLOCK(p);
1315         return (0);
1316 }
1317
1318 bool
1319 thread_suspend_check_needed(void)
1320 {
1321         struct proc *p;
1322         struct thread *td;
1323
1324         td = curthread;
1325         p = td->td_proc;
1326         PROC_LOCK_ASSERT(p, MA_OWNED);
1327         return (P_SHOULDSTOP(p) || ((p->p_flag & P_TRACED) != 0 &&
1328             (td->td_dbgflags & TDB_SUSPEND) != 0));
1329 }
1330
1331 /*
1332  * Called in from locations that can safely check to see
1333  * whether we have to suspend or at least throttle for a
1334  * single-thread event (e.g. fork).
1335  *
1336  * Such locations include userret().
1337  * If the "return_instead" argument is non zero, the thread must be able to
1338  * accept 0 (caller may continue), or 1 (caller must abort) as a result.
1339  *
1340  * The 'return_instead' argument tells the function if it may do a
1341  * thread_exit() or suspend, or whether the caller must abort and back
1342  * out instead.
1343  *
1344  * If the thread that set the single_threading request has set the
1345  * P_SINGLE_EXIT bit in the process flags then this call will never return
1346  * if 'return_instead' is false, but will exit.
1347  *
1348  * P_SINGLE_EXIT | return_instead == 0| return_instead != 0
1349  *---------------+--------------------+---------------------
1350  *       0       | returns 0          |   returns 0 or 1
1351  *               | when ST ends       |   immediately
1352  *---------------+--------------------+---------------------
1353  *       1       | thread exits       |   returns 1
1354  *               |                    |  immediately
1355  * 0 = thread_exit() or suspension ok,
1356  * other = return error instead of stopping the thread.
1357  *
1358  * While a full suspension is under effect, even a single threading
1359  * thread would be suspended if it made this call (but it shouldn't).
1360  * This call should only be made from places where
1361  * thread_exit() would be safe as that may be the outcome unless
1362  * return_instead is set.
1363  */
1364 int
1365 thread_suspend_check(int return_instead)
1366 {
1367         struct thread *td;
1368         struct proc *p;
1369         int wakeup_swapper;
1370
1371         td = curthread;
1372         p = td->td_proc;
1373         mtx_assert(&Giant, MA_NOTOWNED);
1374         PROC_LOCK_ASSERT(p, MA_OWNED);
1375         while (thread_suspend_check_needed()) {
1376                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
1377                         KASSERT(p->p_singlethread != NULL,
1378                             ("singlethread not set"));
1379                         /*
1380                          * The only suspension in action is a
1381                          * single-threading. Single threader need not stop.
1382                          * It is safe to access p->p_singlethread unlocked
1383                          * because it can only be set to our address by us.
1384                          */
1385                         if (p->p_singlethread == td)
1386                                 return (0);     /* Exempt from stopping. */
1387                 }
1388                 if ((p->p_flag & P_SINGLE_EXIT) && return_instead)
1389                         return (EINTR);
1390
1391                 /* Should we goto user boundary if we didn't come from there? */
1392                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1393                     (p->p_flag & P_SINGLE_BOUNDARY) && return_instead)
1394                         return (ERESTART);
1395
1396                 /*
1397                  * Ignore suspend requests if they are deferred.
1398                  */
1399                 if ((td->td_flags & TDF_SBDRY) != 0) {
1400                         KASSERT(return_instead,
1401                             ("TDF_SBDRY set for unsafe thread_suspend_check"));
1402                         KASSERT((td->td_flags & (TDF_SEINTR | TDF_SERESTART)) !=
1403                             (TDF_SEINTR | TDF_SERESTART),
1404                             ("both TDF_SEINTR and TDF_SERESTART"));
1405                         return (TD_SBDRY_INTR(td) ? TD_SBDRY_ERRNO(td) : 0);
1406                 }
1407
1408                 /*
1409                  * If the process is waiting for us to exit,
1410                  * this thread should just suicide.
1411                  * Assumes that P_SINGLE_EXIT implies P_STOPPED_SINGLE.
1412                  */
1413                 if ((p->p_flag & P_SINGLE_EXIT) && (p->p_singlethread != td)) {
1414                         PROC_UNLOCK(p);
1415
1416                         /*
1417                          * Allow Linux emulation layer to do some work
1418                          * before thread suicide.
1419                          */
1420                         if (__predict_false(p->p_sysent->sv_thread_detach != NULL))
1421                                 (p->p_sysent->sv_thread_detach)(td);
1422                         umtx_thread_exit(td);
1423                         kern_thr_exit(td);
1424                         panic("stopped thread did not exit");
1425                 }
1426
1427                 PROC_SLOCK(p);
1428                 thread_stopped(p);
1429                 if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE) {
1430                         if (p->p_numthreads == p->p_suspcount + 1) {
1431                                 thread_lock(p->p_singlethread);
1432                                 wakeup_swapper = thread_unsuspend_one(
1433                                     p->p_singlethread, p, false);
1434                                 if (wakeup_swapper)
1435                                         kick_proc0();
1436                         }
1437                 }
1438                 PROC_UNLOCK(p);
1439                 thread_lock(td);
1440                 /*
1441                  * When a thread suspends, it just
1442                  * gets taken off all queues.
1443                  */
1444                 thread_suspend_one(td);
1445                 if (return_instead == 0) {
1446                         p->p_boundary_count++;
1447                         td->td_flags |= TDF_BOUNDARY;
1448                 }
1449                 PROC_SUNLOCK(p);
1450                 mi_switch(SW_INVOL | SWT_SUSPEND);
1451                 PROC_LOCK(p);
1452         }
1453         return (0);
1454 }
1455
1456 /*
1457  * Check for possible stops and suspensions while executing a
1458  * casueword or similar transiently failing operation.
1459  *
1460  * The sleep argument controls whether the function can handle a stop
1461  * request itself or it should return ERESTART and the request is
1462  * proceed at the kernel/user boundary in ast.
1463  *
1464  * Typically, when retrying due to casueword(9) failure (rv == 1), we
1465  * should handle the stop requests there, with exception of cases when
1466  * the thread owns a kernel resource, for instance busied the umtx
1467  * key, or when functions return immediately if thread_check_susp()
1468  * returned non-zero.  On the other hand, retrying the whole lock
1469  * operation, we better not stop there but delegate the handling to
1470  * ast.
1471  *
1472  * If the request is for thread termination P_SINGLE_EXIT, we cannot
1473  * handle it at all, and simply return EINTR.
1474  */
1475 int
1476 thread_check_susp(struct thread *td, bool sleep)
1477 {
1478         struct proc *p;
1479         int error;
1480
1481         /*
1482          * The check for TDF_NEEDSUSPCHK is racy, but it is enough to
1483          * eventually break the lockstep loop.
1484          */
1485         if ((td->td_flags & TDF_NEEDSUSPCHK) == 0)
1486                 return (0);
1487         error = 0;
1488         p = td->td_proc;
1489         PROC_LOCK(p);
1490         if (p->p_flag & P_SINGLE_EXIT)
1491                 error = EINTR;
1492         else if (P_SHOULDSTOP(p) ||
1493             ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND)))
1494                 error = sleep ? thread_suspend_check(0) : ERESTART;
1495         PROC_UNLOCK(p);
1496         return (error);
1497 }
1498
1499 void
1500 thread_suspend_switch(struct thread *td, struct proc *p)
1501 {
1502
1503         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1504         PROC_LOCK_ASSERT(p, MA_OWNED);
1505         PROC_SLOCK_ASSERT(p, MA_OWNED);
1506         /*
1507          * We implement thread_suspend_one in stages here to avoid
1508          * dropping the proc lock while the thread lock is owned.
1509          */
1510         if (p == td->td_proc) {
1511                 thread_stopped(p);
1512                 p->p_suspcount++;
1513         }
1514         PROC_UNLOCK(p);
1515         thread_lock(td);
1516         td->td_flags &= ~TDF_NEEDSUSPCHK;
1517         TD_SET_SUSPENDED(td);
1518         sched_sleep(td, 0);
1519         PROC_SUNLOCK(p);
1520         DROP_GIANT();
1521         mi_switch(SW_VOL | SWT_SUSPEND);
1522         PICKUP_GIANT();
1523         PROC_LOCK(p);
1524         PROC_SLOCK(p);
1525 }
1526
1527 void
1528 thread_suspend_one(struct thread *td)
1529 {
1530         struct proc *p;
1531
1532         p = td->td_proc;
1533         PROC_SLOCK_ASSERT(p, MA_OWNED);
1534         THREAD_LOCK_ASSERT(td, MA_OWNED);
1535         KASSERT(!TD_IS_SUSPENDED(td), ("already suspended"));
1536         p->p_suspcount++;
1537         td->td_flags &= ~TDF_NEEDSUSPCHK;
1538         TD_SET_SUSPENDED(td);
1539         sched_sleep(td, 0);
1540 }
1541
1542 static int
1543 thread_unsuspend_one(struct thread *td, struct proc *p, bool boundary)
1544 {
1545
1546         THREAD_LOCK_ASSERT(td, MA_OWNED);
1547         KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
1548         TD_CLR_SUSPENDED(td);
1549         td->td_flags &= ~TDF_ALLPROCSUSP;
1550         if (td->td_proc == p) {
1551                 PROC_SLOCK_ASSERT(p, MA_OWNED);
1552                 p->p_suspcount--;
1553                 if (boundary && (td->td_flags & TDF_BOUNDARY) != 0) {
1554                         td->td_flags &= ~TDF_BOUNDARY;
1555                         p->p_boundary_count--;
1556                 }
1557         }
1558         return (setrunnable(td, 0));
1559 }
1560
1561 void
1562 thread_run_flash(struct thread *td)
1563 {
1564         struct proc *p;
1565
1566         p = td->td_proc;
1567         PROC_LOCK_ASSERT(p, MA_OWNED);
1568
1569         if (TD_ON_SLEEPQ(td))
1570                 sleepq_remove_nested(td);
1571         else
1572                 thread_lock(td);
1573
1574         THREAD_LOCK_ASSERT(td, MA_OWNED);
1575         KASSERT(TD_IS_SUSPENDED(td), ("Thread not suspended"));
1576
1577         TD_CLR_SUSPENDED(td);
1578         PROC_SLOCK(p);
1579         MPASS(p->p_suspcount > 0);
1580         p->p_suspcount--;
1581         PROC_SUNLOCK(p);
1582         if (setrunnable(td, 0))
1583                 kick_proc0();
1584 }
1585
1586 /*
1587  * Allow all threads blocked by single threading to continue running.
1588  */
1589 void
1590 thread_unsuspend(struct proc *p)
1591 {
1592         struct thread *td;
1593         int wakeup_swapper;
1594
1595         PROC_LOCK_ASSERT(p, MA_OWNED);
1596         PROC_SLOCK_ASSERT(p, MA_OWNED);
1597         wakeup_swapper = 0;
1598         if (!P_SHOULDSTOP(p)) {
1599                 FOREACH_THREAD_IN_PROC(p, td) {
1600                         thread_lock(td);
1601                         if (TD_IS_SUSPENDED(td)) {
1602                                 wakeup_swapper |= thread_unsuspend_one(td, p,
1603                                     true);
1604                         } else
1605                                 thread_unlock(td);
1606                 }
1607         } else if (P_SHOULDSTOP(p) == P_STOPPED_SINGLE &&
1608             p->p_numthreads == p->p_suspcount) {
1609                 /*
1610                  * Stopping everything also did the job for the single
1611                  * threading request. Now we've downgraded to single-threaded,
1612                  * let it continue.
1613                  */
1614                 if (p->p_singlethread->td_proc == p) {
1615                         thread_lock(p->p_singlethread);
1616                         wakeup_swapper = thread_unsuspend_one(
1617                             p->p_singlethread, p, false);
1618                 }
1619         }
1620         if (wakeup_swapper)
1621                 kick_proc0();
1622 }
1623
1624 /*
1625  * End the single threading mode..
1626  */
1627 void
1628 thread_single_end(struct proc *p, int mode)
1629 {
1630         struct thread *td;
1631         int wakeup_swapper;
1632
1633         KASSERT(mode == SINGLE_EXIT || mode == SINGLE_BOUNDARY ||
1634             mode == SINGLE_ALLPROC || mode == SINGLE_NO_EXIT,
1635             ("invalid mode %d", mode));
1636         PROC_LOCK_ASSERT(p, MA_OWNED);
1637         KASSERT((mode == SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) != 0) ||
1638             (mode != SINGLE_ALLPROC && (p->p_flag & P_TOTAL_STOP) == 0),
1639             ("mode %d does not match P_TOTAL_STOP", mode));
1640         KASSERT(mode == SINGLE_ALLPROC || p->p_singlethread == curthread,
1641             ("thread_single_end from other thread %p %p",
1642             curthread, p->p_singlethread));
1643         KASSERT(mode != SINGLE_BOUNDARY ||
1644             (p->p_flag & P_SINGLE_BOUNDARY) != 0,
1645             ("mis-matched SINGLE_BOUNDARY flags %x", p->p_flag));
1646         p->p_flag &= ~(P_STOPPED_SINGLE | P_SINGLE_EXIT | P_SINGLE_BOUNDARY |
1647             P_TOTAL_STOP);
1648         PROC_SLOCK(p);
1649         p->p_singlethread = NULL;
1650         wakeup_swapper = 0;
1651         /*
1652          * If there are other threads they may now run,
1653          * unless of course there is a blanket 'stop order'
1654          * on the process. The single threader must be allowed
1655          * to continue however as this is a bad place to stop.
1656          */
1657         if (p->p_numthreads != remain_for_mode(mode) && !P_SHOULDSTOP(p)) {
1658                 FOREACH_THREAD_IN_PROC(p, td) {
1659                         thread_lock(td);
1660                         if (TD_IS_SUSPENDED(td)) {
1661                                 wakeup_swapper |= thread_unsuspend_one(td, p,
1662                                     mode == SINGLE_BOUNDARY);
1663                         } else
1664                                 thread_unlock(td);
1665                 }
1666         }
1667         KASSERT(mode != SINGLE_BOUNDARY || p->p_boundary_count == 0,
1668             ("inconsistent boundary count %d", p->p_boundary_count));
1669         PROC_SUNLOCK(p);
1670         if (wakeup_swapper)
1671                 kick_proc0();
1672 }
1673
1674 /*
1675  * Locate a thread by number and return with proc lock held.
1676  *
1677  * thread exit establishes proc -> tidhash lock ordering, but lookup
1678  * takes tidhash first and needs to return locked proc.
1679  *
1680  * The problem is worked around by relying on type-safety of both
1681  * structures and doing the work in 2 steps:
1682  * - tidhash-locked lookup which saves both thread and proc pointers
1683  * - proc-locked verification that the found thread still matches
1684  */
1685 static bool
1686 tdfind_hash(lwpid_t tid, pid_t pid, struct proc **pp, struct thread **tdp)
1687 {
1688 #define RUN_THRESH      16
1689         struct proc *p;
1690         struct thread *td;
1691         int run;
1692         bool locked;
1693
1694         run = 0;
1695         rw_rlock(TIDHASHLOCK(tid));
1696         locked = true;
1697         LIST_FOREACH(td, TIDHASH(tid), td_hash) {
1698                 if (td->td_tid != tid) {
1699                         run++;
1700                         continue;
1701                 }
1702                 p = td->td_proc;
1703                 if (pid != -1 && p->p_pid != pid) {
1704                         td = NULL;
1705                         break;
1706                 }
1707                 if (run > RUN_THRESH) {
1708                         if (rw_try_upgrade(TIDHASHLOCK(tid))) {
1709                                 LIST_REMOVE(td, td_hash);
1710                                 LIST_INSERT_HEAD(TIDHASH(td->td_tid),
1711                                         td, td_hash);
1712                                 rw_wunlock(TIDHASHLOCK(tid));
1713                                 locked = false;
1714                                 break;
1715                         }
1716                 }
1717                 break;
1718         }
1719         if (locked)
1720                 rw_runlock(TIDHASHLOCK(tid));
1721         if (td == NULL)
1722                 return (false);
1723         *pp = p;
1724         *tdp = td;
1725         return (true);
1726 }
1727
1728 struct thread *
1729 tdfind(lwpid_t tid, pid_t pid)
1730 {
1731         struct proc *p;
1732         struct thread *td;
1733
1734         td = curthread;
1735         if (td->td_tid == tid) {
1736                 if (pid != -1 && td->td_proc->p_pid != pid)
1737                         return (NULL);
1738                 PROC_LOCK(td->td_proc);
1739                 return (td);
1740         }
1741
1742         for (;;) {
1743                 if (!tdfind_hash(tid, pid, &p, &td))
1744                         return (NULL);
1745                 PROC_LOCK(p);
1746                 if (td->td_tid != tid) {
1747                         PROC_UNLOCK(p);
1748                         continue;
1749                 }
1750                 if (td->td_proc != p) {
1751                         PROC_UNLOCK(p);
1752                         continue;
1753                 }
1754                 if (p->p_state == PRS_NEW) {
1755                         PROC_UNLOCK(p);
1756                         return (NULL);
1757                 }
1758                 return (td);
1759         }
1760 }
1761
1762 void
1763 tidhash_add(struct thread *td)
1764 {
1765         rw_wlock(TIDHASHLOCK(td->td_tid));
1766         LIST_INSERT_HEAD(TIDHASH(td->td_tid), td, td_hash);
1767         rw_wunlock(TIDHASHLOCK(td->td_tid));
1768 }
1769
1770 void
1771 tidhash_remove(struct thread *td)
1772 {
1773
1774         rw_wlock(TIDHASHLOCK(td->td_tid));
1775         LIST_REMOVE(td, td_hash);
1776         rw_wunlock(TIDHASHLOCK(td->td_tid));
1777 }