kern_thr.c revision 315949
116Salm/*- 216Salm * Copyright (c) 2003, Jeffrey Roberson <jeff@freebsd.org> 316Salm * All rights reserved. 416Salm * 516Salm * Redistribution and use in source and binary forms, with or without 616Salm * modification, are permitted provided that the following conditions 716Salm * are met: 816Salm * 1. Redistributions of source code must retain the above copyright 9 * notice unmodified, this list of conditions, and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27#include <sys/cdefs.h> 28__FBSDID("$FreeBSD: stable/11/sys/kern/kern_thr.c 315949 2017-03-25 13:33:23Z badger $"); 29 30#include "opt_compat.h" 31#include "opt_posix.h" 32#include <sys/param.h> 33#include <sys/kernel.h> 34#include <sys/lock.h> 35#include <sys/mutex.h> 36#include <sys/priv.h> 37#include <sys/proc.h> 38#include <sys/posix4.h> 39#include <sys/ptrace.h> 40#include <sys/racct.h> 41#include <sys/resourcevar.h> 42#include <sys/rwlock.h> 43#include <sys/sched.h> 44#include <sys/sysctl.h> 45#include <sys/smp.h> 46#include <sys/syscallsubr.h> 47#include <sys/sysent.h> 48#include <sys/systm.h> 49#include <sys/sysproto.h> 50#include <sys/signalvar.h> 51#include <sys/sysctl.h> 52#include <sys/ucontext.h> 53#include <sys/thr.h> 54#include <sys/rtprio.h> 55#include <sys/umtx.h> 56#include <sys/limits.h> 57 58#include <vm/vm_domain.h> 59 60#include <machine/frame.h> 61 62#include <security/audit/audit.h> 63 64static SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0, 65 "thread allocation"); 66 67static int max_threads_per_proc = 1500; 68SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW, 69 &max_threads_per_proc, 0, "Limit on threads per proc"); 70 71static int max_threads_hits; 72SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD, 73 &max_threads_hits, 0, "kern.threads.max_threads_per_proc hit count"); 74 75#ifdef COMPAT_FREEBSD32 76 77static inline int 78suword_lwpid(void *addr, lwpid_t lwpid) 79{ 80 int error; 81 82 if (SV_CURPROC_FLAG(SV_LP64)) 83 error = suword(addr, lwpid); 84 else 85 error = suword32(addr, lwpid); 86 return (error); 87} 88 89#else 90#define suword_lwpid suword 91#endif 92 93/* 94 * System call interface. 95 */ 96 97struct thr_create_initthr_args { 98 ucontext_t ctx; 99 long *tid; 100}; 101 102static int 103thr_create_initthr(struct thread *td, void *thunk) 104{ 105 struct thr_create_initthr_args *args; 106 107 /* Copy out the child tid. */ 108 args = thunk; 109 if (args->tid != NULL && suword_lwpid(args->tid, td->td_tid)) 110 return (EFAULT); 111 112 return (set_mcontext(td, &args->ctx.uc_mcontext)); 113} 114 115int 116sys_thr_create(struct thread *td, struct thr_create_args *uap) 117 /* ucontext_t *ctx, long *id, int flags */ 118{ 119 struct thr_create_initthr_args args; 120 int error; 121 122 if ((error = copyin(uap->ctx, &args.ctx, sizeof(args.ctx)))) 123 return (error); 124 args.tid = uap->id; 125 return (thread_create(td, NULL, thr_create_initthr, &args)); 126} 127 128int 129sys_thr_new(struct thread *td, struct thr_new_args *uap) 130 /* struct thr_param * */ 131{ 132 struct thr_param param; 133 int error; 134 135 if (uap->param_size < 0 || uap->param_size > sizeof(param)) 136 return (EINVAL); 137 bzero(¶m, sizeof(param)); 138 if ((error = copyin(uap->param, ¶m, uap->param_size))) 139 return (error); 140 return (kern_thr_new(td, ¶m)); 141} 142 143static int 144thr_new_initthr(struct thread *td, void *thunk) 145{ 146 stack_t stack; 147 struct thr_param *param; 148 149 /* 150 * Here we copy out tid to two places, one for child and one 151 * for parent, because pthread can create a detached thread, 152 * if parent wants to safely access child tid, it has to provide 153 * its storage, because child thread may exit quickly and 154 * memory is freed before parent thread can access it. 155 */ 156 param = thunk; 157 if ((param->child_tid != NULL && 158 suword_lwpid(param->child_tid, td->td_tid)) || 159 (param->parent_tid != NULL && 160 suword_lwpid(param->parent_tid, td->td_tid))) 161 return (EFAULT); 162 163 /* Set up our machine context. */ 164 stack.ss_sp = param->stack_base; 165 stack.ss_size = param->stack_size; 166 /* Set upcall address to user thread entry function. */ 167 cpu_set_upcall(td, param->start_func, param->arg, &stack); 168 /* Setup user TLS address and TLS pointer register. */ 169 return (cpu_set_user_tls(td, param->tls_base)); 170} 171 172int 173kern_thr_new(struct thread *td, struct thr_param *param) 174{ 175 struct rtprio rtp, *rtpp; 176 int error; 177 178 rtpp = NULL; 179 if (param->rtp != 0) { 180 error = copyin(param->rtp, &rtp, sizeof(struct rtprio)); 181 if (error) 182 return (error); 183 rtpp = &rtp; 184 } 185 return (thread_create(td, rtpp, thr_new_initthr, param)); 186} 187 188int 189thread_create(struct thread *td, struct rtprio *rtp, 190 int (*initialize_thread)(struct thread *, void *), void *thunk) 191{ 192 struct thread *newtd; 193 struct proc *p; 194 int error; 195 196 p = td->td_proc; 197 198 if (rtp != NULL) { 199 switch(rtp->type) { 200 case RTP_PRIO_REALTIME: 201 case RTP_PRIO_FIFO: 202 /* Only root can set scheduler policy */ 203 if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0) 204 return (EPERM); 205 if (rtp->prio > RTP_PRIO_MAX) 206 return (EINVAL); 207 break; 208 case RTP_PRIO_NORMAL: 209 rtp->prio = 0; 210 break; 211 default: 212 return (EINVAL); 213 } 214 } 215 216#ifdef RACCT 217 if (racct_enable) { 218 PROC_LOCK(p); 219 error = racct_add(p, RACCT_NTHR, 1); 220 PROC_UNLOCK(p); 221 if (error != 0) 222 return (EPROCLIM); 223 } 224#endif 225 226 /* Initialize our td */ 227 error = kern_thr_alloc(p, 0, &newtd); 228 if (error) 229 goto fail; 230 231 cpu_copy_thread(newtd, td); 232 233 bzero(&newtd->td_startzero, 234 __rangeof(struct thread, td_startzero, td_endzero)); 235 newtd->td_sleeptimo = 0; 236 bcopy(&td->td_startcopy, &newtd->td_startcopy, 237 __rangeof(struct thread, td_startcopy, td_endcopy)); 238 newtd->td_proc = td->td_proc; 239 newtd->td_rb_list = newtd->td_rbp_list = newtd->td_rb_inact = 0; 240 thread_cow_get(newtd, td); 241 242 error = initialize_thread(newtd, thunk); 243 if (error != 0) { 244 thread_cow_free(newtd); 245 thread_free(newtd); 246 goto fail; 247 } 248 249 PROC_LOCK(p); 250 p->p_flag |= P_HADTHREADS; 251 thread_link(newtd, p); 252 bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name)); 253 thread_lock(td); 254 /* let the scheduler know about these things. */ 255 sched_fork_thread(td, newtd); 256 thread_unlock(td); 257 if (P_SHOULDSTOP(p)) 258 newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; 259 if (p->p_ptevents & PTRACE_LWP) 260 newtd->td_dbgflags |= TDB_BORN; 261 262 /* 263 * Copy the existing thread VM policy into the new thread. 264 */ 265 vm_domain_policy_localcopy(&newtd->td_vm_dom_policy, 266 &td->td_vm_dom_policy); 267 268 PROC_UNLOCK(p); 269 270 tidhash_add(newtd); 271 272 thread_lock(newtd); 273 if (rtp != NULL) { 274 if (!(td->td_pri_class == PRI_TIMESHARE && 275 rtp->type == RTP_PRIO_NORMAL)) { 276 rtp_to_pri(rtp, newtd); 277 sched_prio(newtd, newtd->td_user_pri); 278 } /* ignore timesharing class */ 279 } 280 TD_SET_CAN_RUN(newtd); 281 sched_add(newtd, SRQ_BORING); 282 thread_unlock(newtd); 283 284 return (0); 285 286fail: 287#ifdef RACCT 288 if (racct_enable) { 289 PROC_LOCK(p); 290 racct_sub(p, RACCT_NTHR, 1); 291 PROC_UNLOCK(p); 292 } 293#endif 294 return (error); 295} 296 297int 298sys_thr_self(struct thread *td, struct thr_self_args *uap) 299 /* long *id */ 300{ 301 int error; 302 303 error = suword_lwpid(uap->id, (unsigned)td->td_tid); 304 if (error == -1) 305 return (EFAULT); 306 return (0); 307} 308 309int 310sys_thr_exit(struct thread *td, struct thr_exit_args *uap) 311 /* long *state */ 312{ 313 314 umtx_thread_exit(td); 315 316 /* Signal userland that it can free the stack. */ 317 if ((void *)uap->state != NULL) { 318 suword_lwpid(uap->state, 1); 319 kern_umtx_wake(td, uap->state, INT_MAX, 0); 320 } 321 322 return (kern_thr_exit(td)); 323} 324 325int 326kern_thr_exit(struct thread *td) 327{ 328 struct proc *p; 329 330 p = td->td_proc; 331 332 /* 333 * If all of the threads in a process call this routine to 334 * exit (e.g. all threads call pthread_exit()), exactly one 335 * thread should return to the caller to terminate the process 336 * instead of the thread. 337 * 338 * Checking p_numthreads alone is not sufficient since threads 339 * might be committed to terminating while the PROC_LOCK is 340 * dropped in either ptracestop() or while removing this thread 341 * from the tidhash. Instead, the p_pendingexits field holds 342 * the count of threads in either of those states and a thread 343 * is considered the "last" thread if all of the other threads 344 * in a process are already terminating. 345 */ 346 PROC_LOCK(p); 347 if (p->p_numthreads == p->p_pendingexits + 1) { 348 /* 349 * Ignore attempts to shut down last thread in the 350 * proc. This will actually call _exit(2) in the 351 * usermode trampoline when it returns. 352 */ 353 PROC_UNLOCK(p); 354 return (0); 355 } 356 357 p->p_pendingexits++; 358 td->td_dbgflags |= TDB_EXIT; 359 if (p->p_ptevents & PTRACE_LWP) 360 ptracestop(td, SIGTRAP, NULL); 361 PROC_UNLOCK(p); 362 tidhash_remove(td); 363 PROC_LOCK(p); 364 p->p_pendingexits--; 365 366 /* 367 * The check above should prevent all other threads from this 368 * process from exiting while the PROC_LOCK is dropped, so 369 * there must be at least one other thread other than the 370 * current thread. 371 */ 372 KASSERT(p->p_numthreads > 1, ("too few threads")); 373 racct_sub(p, RACCT_NTHR, 1); 374 tdsigcleanup(td); 375 PROC_SLOCK(p); 376 thread_stopped(p); 377 thread_exit(); 378 /* NOTREACHED */ 379} 380 381int 382sys_thr_kill(struct thread *td, struct thr_kill_args *uap) 383 /* long id, int sig */ 384{ 385 ksiginfo_t ksi; 386 struct thread *ttd; 387 struct proc *p; 388 int error; 389 390 p = td->td_proc; 391 ksiginfo_init(&ksi); 392 ksi.ksi_signo = uap->sig; 393 ksi.ksi_code = SI_LWP; 394 ksi.ksi_pid = p->p_pid; 395 ksi.ksi_uid = td->td_ucred->cr_ruid; 396 if (uap->id == -1) { 397 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 398 error = EINVAL; 399 } else { 400 error = ESRCH; 401 PROC_LOCK(p); 402 FOREACH_THREAD_IN_PROC(p, ttd) { 403 if (ttd != td) { 404 error = 0; 405 if (uap->sig == 0) 406 break; 407 tdksignal(ttd, uap->sig, &ksi); 408 } 409 } 410 PROC_UNLOCK(p); 411 } 412 } else { 413 error = 0; 414 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 415 if (ttd == NULL) 416 return (ESRCH); 417 if (uap->sig == 0) 418 ; 419 else if (!_SIG_VALID(uap->sig)) 420 error = EINVAL; 421 else 422 tdksignal(ttd, uap->sig, &ksi); 423 PROC_UNLOCK(ttd->td_proc); 424 } 425 return (error); 426} 427 428int 429sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap) 430 /* pid_t pid, long id, int sig */ 431{ 432 ksiginfo_t ksi; 433 struct thread *ttd; 434 struct proc *p; 435 int error; 436 437 AUDIT_ARG_SIGNUM(uap->sig); 438 439 ksiginfo_init(&ksi); 440 ksi.ksi_signo = uap->sig; 441 ksi.ksi_code = SI_LWP; 442 ksi.ksi_pid = td->td_proc->p_pid; 443 ksi.ksi_uid = td->td_ucred->cr_ruid; 444 if (uap->id == -1) { 445 if ((p = pfind(uap->pid)) == NULL) 446 return (ESRCH); 447 AUDIT_ARG_PROCESS(p); 448 error = p_cansignal(td, p, uap->sig); 449 if (error) { 450 PROC_UNLOCK(p); 451 return (error); 452 } 453 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 454 error = EINVAL; 455 } else { 456 error = ESRCH; 457 FOREACH_THREAD_IN_PROC(p, ttd) { 458 if (ttd != td) { 459 error = 0; 460 if (uap->sig == 0) 461 break; 462 tdksignal(ttd, uap->sig, &ksi); 463 } 464 } 465 } 466 PROC_UNLOCK(p); 467 } else { 468 ttd = tdfind((lwpid_t)uap->id, uap->pid); 469 if (ttd == NULL) 470 return (ESRCH); 471 p = ttd->td_proc; 472 AUDIT_ARG_PROCESS(p); 473 error = p_cansignal(td, p, uap->sig); 474 if (uap->sig == 0) 475 ; 476 else if (!_SIG_VALID(uap->sig)) 477 error = EINVAL; 478 else 479 tdksignal(ttd, uap->sig, &ksi); 480 PROC_UNLOCK(p); 481 } 482 return (error); 483} 484 485int 486sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap) 487 /* const struct timespec *timeout */ 488{ 489 struct timespec ts, *tsp; 490 int error; 491 492 tsp = NULL; 493 if (uap->timeout != NULL) { 494 error = umtx_copyin_timeout(uap->timeout, &ts); 495 if (error != 0) 496 return (error); 497 tsp = &ts; 498 } 499 500 return (kern_thr_suspend(td, tsp)); 501} 502 503int 504kern_thr_suspend(struct thread *td, struct timespec *tsp) 505{ 506 struct proc *p = td->td_proc; 507 struct timeval tv; 508 int error = 0; 509 int timo = 0; 510 511 if (td->td_pflags & TDP_WAKEUP) { 512 td->td_pflags &= ~TDP_WAKEUP; 513 return (0); 514 } 515 516 if (tsp != NULL) { 517 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 518 error = EWOULDBLOCK; 519 else { 520 TIMESPEC_TO_TIMEVAL(&tv, tsp); 521 timo = tvtohz(&tv); 522 } 523 } 524 525 PROC_LOCK(p); 526 if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0) 527 error = msleep((void *)td, &p->p_mtx, 528 PCATCH, "lthr", timo); 529 530 if (td->td_flags & TDF_THRWAKEUP) { 531 thread_lock(td); 532 td->td_flags &= ~TDF_THRWAKEUP; 533 thread_unlock(td); 534 PROC_UNLOCK(p); 535 return (0); 536 } 537 PROC_UNLOCK(p); 538 if (error == EWOULDBLOCK) 539 error = ETIMEDOUT; 540 else if (error == ERESTART) { 541 if (timo != 0) 542 error = EINTR; 543 } 544 return (error); 545} 546 547int 548sys_thr_wake(struct thread *td, struct thr_wake_args *uap) 549 /* long id */ 550{ 551 struct proc *p; 552 struct thread *ttd; 553 554 if (uap->id == td->td_tid) { 555 td->td_pflags |= TDP_WAKEUP; 556 return (0); 557 } 558 559 p = td->td_proc; 560 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 561 if (ttd == NULL) 562 return (ESRCH); 563 thread_lock(ttd); 564 ttd->td_flags |= TDF_THRWAKEUP; 565 thread_unlock(ttd); 566 wakeup((void *)ttd); 567 PROC_UNLOCK(p); 568 return (0); 569} 570 571int 572sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap) 573{ 574 struct proc *p; 575 char name[MAXCOMLEN + 1]; 576 struct thread *ttd; 577 int error; 578 579 error = 0; 580 name[0] = '\0'; 581 if (uap->name != NULL) { 582 error = copyinstr(uap->name, name, sizeof(name), NULL); 583 if (error == ENAMETOOLONG) { 584 error = copyin(uap->name, name, sizeof(name) - 1); 585 name[sizeof(name) - 1] = '\0'; 586 } 587 if (error) 588 return (error); 589 } 590 p = td->td_proc; 591 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 592 if (ttd == NULL) 593 return (ESRCH); 594 strcpy(ttd->td_name, name); 595#ifdef KTR 596 sched_clear_tdname(ttd); 597#endif 598 PROC_UNLOCK(p); 599 return (error); 600} 601 602int 603kern_thr_alloc(struct proc *p, int pages, struct thread **ntd) 604{ 605 606 /* Have race condition but it is cheap. */ 607 if (p->p_numthreads >= max_threads_per_proc) { 608 ++max_threads_hits; 609 return (EPROCLIM); 610 } 611 612 *ntd = thread_alloc(pages); 613 if (*ntd == NULL) 614 return (ENOMEM); 615 616 return (0); 617} 618