1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2003, Jeffrey Roberson <jeff@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 unmodified, this list of conditions, and the following 12 * disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 #include "opt_posix.h" 31 #include "opt_hwpmc_hooks.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/umtxvar.h> 56 #include <sys/limits.h> 57 #ifdef HWPMC_HOOKS 58 #include <sys/pmckern.h> 59 #endif 60 61 #include <machine/frame.h> 62 63 #include <security/audit/audit.h> 64 65 static SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 66 "thread allocation"); 67 68 int max_threads_per_proc = 1500; 69 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW, 70 &max_threads_per_proc, 0, "Limit on threads per proc"); 71 72 static int max_threads_hits; 73 SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD, 74 &max_threads_hits, 0, "kern.threads.max_threads_per_proc hit count"); 75 76 #ifdef COMPAT_FREEBSD32 77 78 static inline int 79 suword_lwpid(void *addr, lwpid_t lwpid) 80 { 81 int error; 82 83 if (SV_CURPROC_FLAG(SV_LP64)) 84 error = suword(addr, lwpid); 85 else 86 error = suword32(addr, lwpid); 87 return (error); 88 } 89 90 #else 91 #define suword_lwpid suword 92 #endif 93 94 /* 95 * System call interface. 96 */ 97 98 struct thr_create_initthr_args { 99 ucontext_t ctx; 100 long *tid; 101 }; 102 103 static int 104 thr_create_initthr(struct thread *td, void *thunk) 105 { 106 struct thr_create_initthr_args *args; 107 108 /* Copy out the child tid. */ 109 args = thunk; 110 if (args->tid != NULL && suword_lwpid(args->tid, td->td_tid)) 111 return (EFAULT); 112 113 return (set_mcontext(td, &args->ctx.uc_mcontext)); 114 } 115 116 int 117 sys_thr_create(struct thread *td, struct thr_create_args *uap) 118 /* ucontext_t *ctx, long *id, int flags */ 119 { 120 struct thr_create_initthr_args args; 121 int error; 122 123 if ((error = copyin(uap->ctx, &args.ctx, sizeof(args.ctx)))) 124 return (error); 125 args.tid = uap->id; 126 return (thread_create(td, NULL, thr_create_initthr, &args)); 127 } 128 129 int 130 sys_thr_new(struct thread *td, struct thr_new_args *uap) 131 /* struct thr_param * */ 132 { 133 struct thr_param param; 134 int error; 135 136 if (uap->param_size < 0 || uap->param_size > sizeof(param)) 137 return (EINVAL); 138 bzero(¶m, sizeof(param)); 139 if ((error = copyin(uap->param, ¶m, uap->param_size))) 140 return (error); 141 return (kern_thr_new(td, ¶m)); 142 } 143 144 static int 145 thr_new_initthr(struct thread *td, void *thunk) 146 { 147 stack_t stack; 148 struct thr_param *param; 149 150 /* 151 * Here we copy out tid to two places, one for child and one 152 * for parent, because pthread can create a detached thread, 153 * if parent wants to safely access child tid, it has to provide 154 * its storage, because child thread may exit quickly and 155 * memory is freed before parent thread can access it. 156 */ 157 param = thunk; 158 if ((param->child_tid != NULL && 159 suword_lwpid(param->child_tid, td->td_tid)) || 160 (param->parent_tid != NULL && 161 suword_lwpid(param->parent_tid, td->td_tid))) 162 return (EFAULT); 163 164 /* Set up our machine context. */ 165 stack.ss_sp = param->stack_base; 166 stack.ss_size = param->stack_size; 167 /* Set upcall address to user thread entry function. */ 168 cpu_set_upcall(td, param->start_func, param->arg, &stack); 169 /* Setup user TLS address and TLS pointer register. */ 170 return (cpu_set_user_tls(td, param->tls_base)); 171 } 172 173 int 174 kern_thr_new(struct thread *td, struct thr_param *param) 175 { 176 struct rtprio rtp, *rtpp; 177 int error; 178 179 rtpp = NULL; 180 if (param->rtp != 0) { 181 error = copyin(param->rtp, &rtp, sizeof(struct rtprio)); 182 if (error) 183 return (error); 184 rtpp = &rtp; 185 } 186 return (thread_create(td, rtpp, thr_new_initthr, param)); 187 } 188 189 int 190 thread_create(struct thread *td, struct rtprio *rtp, 191 int (*initialize_thread)(struct thread *, void *), void *thunk) 192 { 193 struct thread *newtd; 194 struct proc *p; 195 int error; 196 197 p = td->td_proc; 198 199 if (rtp != NULL) { 200 switch(rtp->type) { 201 case RTP_PRIO_REALTIME: 202 case RTP_PRIO_FIFO: 203 /* Only root can set scheduler policy */ 204 if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0) 205 return (EPERM); 206 if (rtp->prio > RTP_PRIO_MAX) 207 return (EINVAL); 208 break; 209 case RTP_PRIO_NORMAL: 210 rtp->prio = 0; 211 break; 212 default: 213 return (EINVAL); 214 } 215 } 216 217 #ifdef RACCT 218 if (racct_enable) { 219 PROC_LOCK(p); 220 error = racct_add(p, RACCT_NTHR, 1); 221 PROC_UNLOCK(p); 222 if (error != 0) 223 return (EPROCLIM); 224 } 225 #endif 226 227 /* Initialize our td */ 228 error = kern_thr_alloc(p, 0, &newtd); 229 if (error) 230 goto fail; 231 232 bzero(&newtd->td_startzero, 233 __rangeof(struct thread, td_startzero, td_endzero)); 234 bcopy(&td->td_startcopy, &newtd->td_startcopy, 235 __rangeof(struct thread, td_startcopy, td_endcopy)); 236 newtd->td_proc = td->td_proc; 237 newtd->td_rb_list = newtd->td_rbp_list = newtd->td_rb_inact = 0; 238 thread_cow_get(newtd, td); 239 240 cpu_copy_thread(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 ast_sched(newtd, TDA_SUSPEND); 259 if (p->p_ptevents & PTRACE_LWP) 260 newtd->td_dbgflags |= TDB_BORN; 261 262 PROC_UNLOCK(p); 263 #ifdef HWPMC_HOOKS 264 if (PMC_PROC_IS_USING_PMCS(p)) 265 PMC_CALL_HOOK(newtd, PMC_FN_THR_CREATE, NULL); 266 else if (PMC_SYSTEM_SAMPLING_ACTIVE()) 267 PMC_CALL_HOOK_UNLOCKED(newtd, PMC_FN_THR_CREATE_LOG, NULL); 268 #endif 269 270 tidhash_add(newtd); 271 272 /* ignore timesharing class */ 273 if (rtp != NULL && !(td->td_pri_class == PRI_TIMESHARE && 274 rtp->type == RTP_PRIO_NORMAL)) 275 rtp_to_pri(rtp, newtd); 276 277 thread_lock(newtd); 278 TD_SET_CAN_RUN(newtd); 279 sched_add(newtd, SRQ_BORING); 280 281 return (0); 282 283 fail: 284 #ifdef RACCT 285 if (racct_enable) { 286 PROC_LOCK(p); 287 racct_sub(p, RACCT_NTHR, 1); 288 PROC_UNLOCK(p); 289 } 290 #endif 291 return (error); 292 } 293 294 int 295 sys_thr_self(struct thread *td, struct thr_self_args *uap) 296 /* long *id */ 297 { 298 int error; 299 300 error = suword_lwpid(uap->id, (unsigned)td->td_tid); 301 if (error == -1) 302 return (EFAULT); 303 return (0); 304 } 305 306 int 307 sys_thr_exit(struct thread *td, struct thr_exit_args *uap) 308 /* long *state */ 309 { 310 311 umtx_thread_exit(td); 312 313 /* Signal userland that it can free the stack. */ 314 if ((void *)uap->state != NULL) { 315 suword_lwpid(uap->state, 1); 316 kern_umtx_wake(td, uap->state, INT_MAX, 0); 317 } 318 319 return (kern_thr_exit(td)); 320 } 321 322 int 323 kern_thr_exit(struct thread *td) 324 { 325 struct proc *p; 326 327 p = td->td_proc; 328 329 /* 330 * If all of the threads in a process call this routine to 331 * exit (e.g. all threads call pthread_exit()), exactly one 332 * thread should return to the caller to terminate the process 333 * instead of the thread. 334 * 335 * Checking p_numthreads alone is not sufficient since threads 336 * might be committed to terminating while the PROC_LOCK is 337 * dropped in either ptracestop() or while removing this thread 338 * from the tidhash. Instead, the p_pendingexits field holds 339 * the count of threads in either of those states and a thread 340 * is considered the "last" thread if all of the other threads 341 * in a process are already terminating. 342 */ 343 PROC_LOCK(p); 344 if (p->p_numthreads == p->p_pendingexits + 1) { 345 /* 346 * Ignore attempts to shut down last thread in the 347 * proc. This will actually call _exit(2) in the 348 * usermode trampoline when it returns. 349 */ 350 PROC_UNLOCK(p); 351 return (0); 352 } 353 354 if (p->p_sysent->sv_ontdexit != NULL) 355 p->p_sysent->sv_ontdexit(td); 356 357 td->td_dbgflags |= TDB_EXIT; 358 if (p->p_ptevents & PTRACE_LWP) { 359 p->p_pendingexits++; 360 ptracestop(td, SIGTRAP, NULL); 361 p->p_pendingexits--; 362 } 363 tidhash_remove(td); 364 365 /* 366 * The check above should prevent all other threads from this 367 * process from exiting while the PROC_LOCK is dropped, so 368 * there must be at least one other thread other than the 369 * current thread. 370 */ 371 KASSERT(p->p_numthreads > 1, ("too few threads")); 372 racct_sub(p, RACCT_NTHR, 1); 373 tdsigcleanup(td); 374 375 #ifdef AUDIT 376 AUDIT_SYSCALL_EXIT(0, td); 377 #endif 378 379 PROC_SLOCK(p); 380 thread_stopped(p); 381 thread_exit(); 382 /* NOTREACHED */ 383 } 384 385 int 386 sys_thr_kill(struct thread *td, struct thr_kill_args *uap) 387 /* long id, int sig */ 388 { 389 ksiginfo_t ksi; 390 struct thread *ttd; 391 struct proc *p; 392 int error; 393 394 p = td->td_proc; 395 ksiginfo_init(&ksi); 396 ksi.ksi_signo = uap->sig; 397 ksi.ksi_code = SI_LWP; 398 ksi.ksi_pid = p->p_pid; 399 ksi.ksi_uid = td->td_ucred->cr_ruid; 400 if (uap->id == -1) { 401 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 402 error = EINVAL; 403 } else { 404 error = ESRCH; 405 PROC_LOCK(p); 406 FOREACH_THREAD_IN_PROC(p, ttd) { 407 if (ttd != td) { 408 error = 0; 409 if (uap->sig == 0) 410 break; 411 tdksignal(ttd, uap->sig, &ksi); 412 } 413 } 414 PROC_UNLOCK(p); 415 } 416 } else { 417 error = 0; 418 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 419 if (ttd == NULL) 420 return (ESRCH); 421 if (uap->sig == 0) 422 ; 423 else if (!_SIG_VALID(uap->sig)) 424 error = EINVAL; 425 else 426 tdksignal(ttd, uap->sig, &ksi); 427 PROC_UNLOCK(ttd->td_proc); 428 } 429 return (error); 430 } 431 432 int 433 sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap) 434 /* pid_t pid, long id, int sig */ 435 { 436 ksiginfo_t ksi; 437 struct thread *ttd; 438 struct proc *p; 439 int error; 440 441 AUDIT_ARG_SIGNUM(uap->sig); 442 443 ksiginfo_init(&ksi); 444 ksi.ksi_signo = uap->sig; 445 ksi.ksi_code = SI_LWP; 446 ksi.ksi_pid = td->td_proc->p_pid; 447 ksi.ksi_uid = td->td_ucred->cr_ruid; 448 if (uap->id == -1) { 449 if ((p = pfind(uap->pid)) == NULL) 450 return (ESRCH); 451 AUDIT_ARG_PROCESS(p); 452 error = p_cansignal(td, p, uap->sig); 453 if (error) { 454 PROC_UNLOCK(p); 455 return (error); 456 } 457 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 458 error = EINVAL; 459 } else { 460 error = ESRCH; 461 FOREACH_THREAD_IN_PROC(p, ttd) { 462 if (ttd != td) { 463 error = 0; 464 if (uap->sig == 0) 465 break; 466 tdksignal(ttd, uap->sig, &ksi); 467 } 468 } 469 } 470 PROC_UNLOCK(p); 471 } else { 472 ttd = tdfind((lwpid_t)uap->id, uap->pid); 473 if (ttd == NULL) 474 return (ESRCH); 475 p = ttd->td_proc; 476 AUDIT_ARG_PROCESS(p); 477 error = p_cansignal(td, p, uap->sig); 478 if (uap->sig == 0) 479 ; 480 else if (!_SIG_VALID(uap->sig)) 481 error = EINVAL; 482 else 483 tdksignal(ttd, uap->sig, &ksi); 484 PROC_UNLOCK(p); 485 } 486 return (error); 487 } 488 489 int 490 sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap) 491 /* const struct timespec *timeout */ 492 { 493 struct timespec ts, *tsp; 494 int error; 495 496 tsp = NULL; 497 if (uap->timeout != NULL) { 498 error = umtx_copyin_timeout(uap->timeout, &ts); 499 if (error != 0) 500 return (error); 501 tsp = &ts; 502 } 503 504 return (kern_thr_suspend(td, tsp)); 505 } 506 507 int 508 kern_thr_suspend(struct thread *td, struct timespec *tsp) 509 { 510 struct proc *p = td->td_proc; 511 struct timeval tv; 512 int error = 0; 513 int timo = 0; 514 515 if (td->td_pflags & TDP_WAKEUP) { 516 td->td_pflags &= ~TDP_WAKEUP; 517 return (0); 518 } 519 520 if (tsp != NULL) { 521 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 522 error = EWOULDBLOCK; 523 else { 524 TIMESPEC_TO_TIMEVAL(&tv, tsp); 525 timo = tvtohz(&tv); 526 } 527 } 528 529 PROC_LOCK(p); 530 if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0) 531 error = msleep((void *)td, &p->p_mtx, 532 PCATCH, "lthr", timo); 533 534 if (td->td_flags & TDF_THRWAKEUP) { 535 thread_lock(td); 536 td->td_flags &= ~TDF_THRWAKEUP; 537 thread_unlock(td); 538 PROC_UNLOCK(p); 539 return (0); 540 } 541 PROC_UNLOCK(p); 542 if (error == EWOULDBLOCK) 543 error = ETIMEDOUT; 544 else if (error == ERESTART) { 545 if (timo != 0) 546 error = EINTR; 547 } 548 return (error); 549 } 550 551 int 552 sys_thr_wake(struct thread *td, struct thr_wake_args *uap) 553 /* long id */ 554 { 555 struct proc *p; 556 struct thread *ttd; 557 558 if (uap->id == td->td_tid) { 559 td->td_pflags |= TDP_WAKEUP; 560 return (0); 561 } 562 563 p = td->td_proc; 564 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 565 if (ttd == NULL) 566 return (ESRCH); 567 thread_lock(ttd); 568 ttd->td_flags |= TDF_THRWAKEUP; 569 thread_unlock(ttd); 570 wakeup((void *)ttd); 571 PROC_UNLOCK(p); 572 return (0); 573 } 574 575 int 576 sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap) 577 { 578 struct proc *p; 579 char name[MAXCOMLEN + 1]; 580 struct thread *ttd; 581 int error; 582 583 error = 0; 584 name[0] = '\0'; 585 if (uap->name != NULL) { 586 error = copyinstr(uap->name, name, sizeof(name), NULL); 587 if (error == ENAMETOOLONG) { 588 error = copyin(uap->name, name, sizeof(name) - 1); 589 name[sizeof(name) - 1] = '\0'; 590 } 591 if (error) 592 return (error); 593 } 594 p = td->td_proc; 595 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 596 if (ttd == NULL) 597 return (ESRCH); 598 strcpy(ttd->td_name, name); 599 #ifdef HWPMC_HOOKS 600 if (PMC_PROC_IS_USING_PMCS(p) || PMC_SYSTEM_SAMPLING_ACTIVE()) 601 PMC_CALL_HOOK_UNLOCKED(ttd, PMC_FN_THR_CREATE_LOG, NULL); 602 #endif 603 #ifdef KTR 604 sched_clear_tdname(ttd); 605 #endif 606 PROC_UNLOCK(p); 607 return (error); 608 } 609 610 int 611 kern_thr_alloc(struct proc *p, int pages, struct thread **ntd) 612 { 613 614 /* Have race condition but it is cheap. */ 615 if (p->p_numthreads >= max_threads_per_proc) { 616 ++max_threads_hits; 617 return (EPROCLIM); 618 } 619 620 *ntd = thread_alloc(pages); 621 if (*ntd == NULL) 622 return (ENOMEM); 623 624 return (0); 625 } 626