kern_thr.c revision 304188
1/*- 2 * Copyright (c) 2003, Jeffrey Roberson <jeff@freebsd.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice 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/10/sys/kern/kern_thr.c 304188 2016-08-15 21:10:41Z jhb $"); 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 <machine/frame.h> 59 60#include <security/audit/audit.h> 61 62static SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0, 63 "thread allocation"); 64 65static int max_threads_per_proc = 1500; 66SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW, 67 &max_threads_per_proc, 0, "Limit on threads per proc"); 68 69static int max_threads_hits; 70SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD, 71 &max_threads_hits, 0, "kern.threads.max_threads_per_proc hit count"); 72 73#ifdef COMPAT_FREEBSD32 74 75static inline int 76suword_lwpid(void *addr, lwpid_t lwpid) 77{ 78 int error; 79 80 if (SV_CURPROC_FLAG(SV_LP64)) 81 error = suword(addr, lwpid); 82 else 83 error = suword32(addr, lwpid); 84 return (error); 85} 86 87#else 88#define suword_lwpid suword 89#endif 90 91/* 92 * System call interface. 93 */ 94 95struct thr_create_initthr_args { 96 ucontext_t ctx; 97 long *tid; 98}; 99 100static int 101thr_create_initthr(struct thread *td, void *thunk) 102{ 103 struct thr_create_initthr_args *args; 104 105 /* Copy out the child tid. */ 106 args = thunk; 107 if (args->tid != NULL && suword_lwpid(args->tid, td->td_tid)) 108 return (EFAULT); 109 110 return (set_mcontext(td, &args->ctx.uc_mcontext)); 111} 112 113int 114sys_thr_create(struct thread *td, struct thr_create_args *uap) 115 /* ucontext_t *ctx, long *id, int flags */ 116{ 117 struct thr_create_initthr_args args; 118 int error; 119 120 if ((error = copyin(uap->ctx, &args.ctx, sizeof(args.ctx)))) 121 return (error); 122 args.tid = uap->id; 123 return (thread_create(td, NULL, thr_create_initthr, &args)); 124} 125 126int 127sys_thr_new(struct thread *td, struct thr_new_args *uap) 128 /* struct thr_param * */ 129{ 130 struct thr_param param; 131 int error; 132 133 if (uap->param_size < 0 || uap->param_size > sizeof(param)) 134 return (EINVAL); 135 bzero(¶m, sizeof(param)); 136 if ((error = copyin(uap->param, ¶m, uap->param_size))) 137 return (error); 138 return (kern_thr_new(td, ¶m)); 139} 140 141static int 142thr_new_initthr(struct thread *td, void *thunk) 143{ 144 stack_t stack; 145 struct thr_param *param; 146 147 /* 148 * Here we copy out tid to two places, one for child and one 149 * for parent, because pthread can create a detached thread, 150 * if parent wants to safely access child tid, it has to provide 151 * its storage, because child thread may exit quickly and 152 * memory is freed before parent thread can access it. 153 */ 154 param = thunk; 155 if ((param->child_tid != NULL && 156 suword_lwpid(param->child_tid, td->td_tid)) || 157 (param->parent_tid != NULL && 158 suword_lwpid(param->parent_tid, td->td_tid))) 159 return (EFAULT); 160 161 /* Set up our machine context. */ 162 stack.ss_sp = param->stack_base; 163 stack.ss_size = param->stack_size; 164 /* Set upcall address to user thread entry function. */ 165 cpu_set_upcall_kse(td, param->start_func, param->arg, &stack); 166 /* Setup user TLS address and TLS pointer register. */ 167 return (cpu_set_user_tls(td, param->tls_base)); 168} 169 170int 171kern_thr_new(struct thread *td, struct thr_param *param) 172{ 173 struct rtprio rtp, *rtpp; 174 int error; 175 176 rtpp = NULL; 177 if (param->rtp != 0) { 178 error = copyin(param->rtp, &rtp, sizeof(struct rtprio)); 179 if (error) 180 return (error); 181 rtpp = &rtp; 182 } 183 return (thread_create(td, rtpp, thr_new_initthr, param)); 184} 185 186int 187thread_create(struct thread *td, struct rtprio *rtp, 188 int (*initialize_thread)(struct thread *, void *), void *thunk) 189{ 190 struct thread *newtd; 191 struct proc *p; 192 int error; 193 194 p = td->td_proc; 195 196 if (rtp != NULL) { 197 switch(rtp->type) { 198 case RTP_PRIO_REALTIME: 199 case RTP_PRIO_FIFO: 200 /* Only root can set scheduler policy */ 201 if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0) 202 return (EPERM); 203 if (rtp->prio > RTP_PRIO_MAX) 204 return (EINVAL); 205 break; 206 case RTP_PRIO_NORMAL: 207 rtp->prio = 0; 208 break; 209 default: 210 return (EINVAL); 211 } 212 } 213 214#ifdef RACCT 215 PROC_LOCK(td->td_proc); 216 error = racct_add(p, RACCT_NTHR, 1); 217 PROC_UNLOCK(td->td_proc); 218 if (error != 0) 219 return (EPROCLIM); 220#endif 221 222 /* Initialize our td */ 223 error = kern_thr_alloc(p, 0, &newtd); 224 if (error) 225 goto fail; 226 227 cpu_set_upcall(newtd, td); 228 229 bzero(&newtd->td_startzero, 230 __rangeof(struct thread, td_startzero, td_endzero)); 231 newtd->td_su = NULL; 232 bcopy(&td->td_startcopy, &newtd->td_startcopy, 233 __rangeof(struct thread, td_startcopy, td_endcopy)); 234 newtd->td_proc = td->td_proc; 235 newtd->td_ucred = crhold(td->td_ucred); 236 newtd->td_dbg_sc_code = td->td_dbg_sc_code; 237 newtd->td_dbg_sc_narg = td->td_dbg_sc_narg; 238 239 error = initialize_thread(newtd, thunk); 240 if (error != 0) { 241 thread_free(newtd); 242 crfree(td->td_ucred); 243 goto fail; 244 } 245 246 PROC_LOCK(td->td_proc); 247 td->td_proc->p_flag |= P_HADTHREADS; 248 thread_link(newtd, p); 249 bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name)); 250 thread_lock(td); 251 /* let the scheduler know about these things. */ 252 sched_fork_thread(td, newtd); 253 thread_unlock(td); 254 if (P_SHOULDSTOP(p)) 255 newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; 256 if (p->p_ptevents & PTRACE_LWP) 257 newtd->td_dbgflags |= TDB_BORN; 258 PROC_UNLOCK(p); 259 260 tidhash_add(newtd); 261 262 thread_lock(newtd); 263 if (rtp != NULL) { 264 if (!(td->td_pri_class == PRI_TIMESHARE && 265 rtp->type == RTP_PRIO_NORMAL)) { 266 rtp_to_pri(rtp, newtd); 267 sched_prio(newtd, newtd->td_user_pri); 268 } /* ignore timesharing class */ 269 } 270 TD_SET_CAN_RUN(newtd); 271 sched_add(newtd, SRQ_BORING); 272 thread_unlock(newtd); 273 274 return (0); 275 276fail: 277#ifdef RACCT 278 if (racct_enable) { 279 PROC_LOCK(p); 280 racct_sub(p, RACCT_NTHR, 1); 281 PROC_UNLOCK(p); 282 } 283#endif 284 return (error); 285} 286 287int 288sys_thr_self(struct thread *td, struct thr_self_args *uap) 289 /* long *id */ 290{ 291 int error; 292 293 error = suword_lwpid(uap->id, (unsigned)td->td_tid); 294 if (error == -1) 295 return (EFAULT); 296 return (0); 297} 298 299int 300sys_thr_exit(struct thread *td, struct thr_exit_args *uap) 301 /* long *state */ 302{ 303 304 /* Signal userland that it can free the stack. */ 305 if ((void *)uap->state != NULL) { 306 suword_lwpid(uap->state, 1); 307 kern_umtx_wake(td, uap->state, INT_MAX, 0); 308 } 309 310 return (kern_thr_exit(td)); 311} 312 313int 314kern_thr_exit(struct thread *td) 315{ 316 struct proc *p; 317 318 p = td->td_proc; 319 320 /* 321 * If all of the threads in a process call this routine to 322 * exit (e.g. all threads call pthread_exit()), exactly one 323 * thread should return to the caller to terminate the process 324 * instead of the thread. 325 * 326 * Checking p_numthreads alone is not sufficient since threads 327 * might be committed to terminating while the PROC_LOCK is 328 * dropped in either ptracestop() or while removing this thread 329 * from the tidhash. Instead, the p_pendingexits field holds 330 * the count of threads in either of those states and a thread 331 * is considered the "last" thread if all of the other threads 332 * in a process are already terminating. 333 */ 334 PROC_LOCK(p); 335 if (p->p_numthreads == p->p_pendingexits + 1) { 336 /* 337 * Ignore attempts to shut down last thread in the 338 * proc. This will actually call _exit(2) in the 339 * usermode trampoline when it returns. 340 */ 341 PROC_UNLOCK(p); 342 return (0); 343 } 344 345 p->p_pendingexits++; 346 td->td_dbgflags |= TDB_EXIT; 347 if (p->p_ptevents & PTRACE_LWP) 348 ptracestop(td, SIGTRAP); 349 PROC_UNLOCK(p); 350 tidhash_remove(td); 351 PROC_LOCK(p); 352 p->p_pendingexits--; 353 354 /* 355 * The check above should prevent all other threads from this 356 * process from exiting while the PROC_LOCK is dropped, so 357 * there must be at least one other thread other than the 358 * current thread. 359 */ 360 KASSERT(p->p_numthreads > 1, ("too few threads")); 361 racct_sub(p, RACCT_NTHR, 1); 362 tdsigcleanup(td); 363 umtx_thread_exit(td); 364 PROC_SLOCK(p); 365 thread_stopped(p); 366 thread_exit(); 367 /* NOTREACHED */ 368} 369 370int 371sys_thr_kill(struct thread *td, struct thr_kill_args *uap) 372 /* long id, int sig */ 373{ 374 ksiginfo_t ksi; 375 struct thread *ttd; 376 struct proc *p; 377 int error; 378 379 p = td->td_proc; 380 ksiginfo_init(&ksi); 381 ksi.ksi_signo = uap->sig; 382 ksi.ksi_code = SI_LWP; 383 ksi.ksi_pid = p->p_pid; 384 ksi.ksi_uid = td->td_ucred->cr_ruid; 385 if (uap->id == -1) { 386 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 387 error = EINVAL; 388 } else { 389 error = ESRCH; 390 PROC_LOCK(p); 391 FOREACH_THREAD_IN_PROC(p, ttd) { 392 if (ttd != td) { 393 error = 0; 394 if (uap->sig == 0) 395 break; 396 tdksignal(ttd, uap->sig, &ksi); 397 } 398 } 399 PROC_UNLOCK(p); 400 } 401 } else { 402 error = 0; 403 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 404 if (ttd == NULL) 405 return (ESRCH); 406 if (uap->sig == 0) 407 ; 408 else if (!_SIG_VALID(uap->sig)) 409 error = EINVAL; 410 else 411 tdksignal(ttd, uap->sig, &ksi); 412 PROC_UNLOCK(ttd->td_proc); 413 } 414 return (error); 415} 416 417int 418sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap) 419 /* pid_t pid, long id, int sig */ 420{ 421 ksiginfo_t ksi; 422 struct thread *ttd; 423 struct proc *p; 424 int error; 425 426 AUDIT_ARG_SIGNUM(uap->sig); 427 428 ksiginfo_init(&ksi); 429 ksi.ksi_signo = uap->sig; 430 ksi.ksi_code = SI_LWP; 431 ksi.ksi_pid = td->td_proc->p_pid; 432 ksi.ksi_uid = td->td_ucred->cr_ruid; 433 if (uap->id == -1) { 434 if ((p = pfind(uap->pid)) == NULL) 435 return (ESRCH); 436 AUDIT_ARG_PROCESS(p); 437 error = p_cansignal(td, p, uap->sig); 438 if (error) { 439 PROC_UNLOCK(p); 440 return (error); 441 } 442 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 443 error = EINVAL; 444 } else { 445 error = ESRCH; 446 FOREACH_THREAD_IN_PROC(p, ttd) { 447 if (ttd != td) { 448 error = 0; 449 if (uap->sig == 0) 450 break; 451 tdksignal(ttd, uap->sig, &ksi); 452 } 453 } 454 } 455 PROC_UNLOCK(p); 456 } else { 457 ttd = tdfind((lwpid_t)uap->id, uap->pid); 458 if (ttd == NULL) 459 return (ESRCH); 460 p = ttd->td_proc; 461 AUDIT_ARG_PROCESS(p); 462 error = p_cansignal(td, p, uap->sig); 463 if (uap->sig == 0) 464 ; 465 else if (!_SIG_VALID(uap->sig)) 466 error = EINVAL; 467 else 468 tdksignal(ttd, uap->sig, &ksi); 469 PROC_UNLOCK(p); 470 } 471 return (error); 472} 473 474int 475sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap) 476 /* const struct timespec *timeout */ 477{ 478 struct timespec ts, *tsp; 479 int error; 480 481 tsp = NULL; 482 if (uap->timeout != NULL) { 483 error = umtx_copyin_timeout(uap->timeout, &ts); 484 if (error != 0) 485 return (error); 486 tsp = &ts; 487 } 488 489 return (kern_thr_suspend(td, tsp)); 490} 491 492int 493kern_thr_suspend(struct thread *td, struct timespec *tsp) 494{ 495 struct proc *p = td->td_proc; 496 struct timeval tv; 497 int error = 0; 498 int timo = 0; 499 500 if (td->td_pflags & TDP_WAKEUP) { 501 td->td_pflags &= ~TDP_WAKEUP; 502 return (0); 503 } 504 505 if (tsp != NULL) { 506 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 507 error = EWOULDBLOCK; 508 else { 509 TIMESPEC_TO_TIMEVAL(&tv, tsp); 510 timo = tvtohz(&tv); 511 } 512 } 513 514 PROC_LOCK(p); 515 if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0) 516 error = msleep((void *)td, &p->p_mtx, 517 PCATCH, "lthr", timo); 518 519 if (td->td_flags & TDF_THRWAKEUP) { 520 thread_lock(td); 521 td->td_flags &= ~TDF_THRWAKEUP; 522 thread_unlock(td); 523 PROC_UNLOCK(p); 524 return (0); 525 } 526 PROC_UNLOCK(p); 527 if (error == EWOULDBLOCK) 528 error = ETIMEDOUT; 529 else if (error == ERESTART) { 530 if (timo != 0) 531 error = EINTR; 532 } 533 return (error); 534} 535 536int 537sys_thr_wake(struct thread *td, struct thr_wake_args *uap) 538 /* long id */ 539{ 540 struct proc *p; 541 struct thread *ttd; 542 543 if (uap->id == td->td_tid) { 544 td->td_pflags |= TDP_WAKEUP; 545 return (0); 546 } 547 548 p = td->td_proc; 549 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 550 if (ttd == NULL) 551 return (ESRCH); 552 thread_lock(ttd); 553 ttd->td_flags |= TDF_THRWAKEUP; 554 thread_unlock(ttd); 555 wakeup((void *)ttd); 556 PROC_UNLOCK(p); 557 return (0); 558} 559 560int 561sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap) 562{ 563 struct proc *p; 564 char name[MAXCOMLEN + 1]; 565 struct thread *ttd; 566 int error; 567 568 error = 0; 569 name[0] = '\0'; 570 if (uap->name != NULL) { 571 error = copyinstr(uap->name, name, sizeof(name), 572 NULL); 573 if (error) 574 return (error); 575 } 576 p = td->td_proc; 577 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 578 if (ttd == NULL) 579 return (ESRCH); 580 strcpy(ttd->td_name, name); 581#ifdef KTR 582 sched_clear_tdname(ttd); 583#endif 584 PROC_UNLOCK(p); 585 return (error); 586} 587 588int 589kern_thr_alloc(struct proc *p, int pages, struct thread **ntd) 590{ 591 592 /* Have race condition but it is cheap. */ 593 if (p->p_numthreads >= max_threads_per_proc) { 594 ++max_threads_hits; 595 return (EPROCLIM); 596 } 597 598 *ntd = thread_alloc(pages); 599 if (*ntd == NULL) 600 return (ENOMEM); 601 602 return (0); 603} 604