kern_thr.c revision 304017
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 304017 2016-08-12 19:43:06Z 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/racct.h> 40#include <sys/resourcevar.h> 41#include <sys/rwlock.h> 42#include <sys/sched.h> 43#include <sys/sysctl.h> 44#include <sys/smp.h> 45#include <sys/syscallsubr.h> 46#include <sys/sysent.h> 47#include <sys/systm.h> 48#include <sys/sysproto.h> 49#include <sys/signalvar.h> 50#include <sys/sysctl.h> 51#include <sys/ucontext.h> 52#include <sys/thr.h> 53#include <sys/rtprio.h> 54#include <sys/umtx.h> 55#include <sys/limits.h> 56 57#include <machine/frame.h> 58 59#include <security/audit/audit.h> 60 61static SYSCTL_NODE(_kern, OID_AUTO, threads, CTLFLAG_RW, 0, 62 "thread allocation"); 63 64static int max_threads_per_proc = 1500; 65SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_per_proc, CTLFLAG_RW, 66 &max_threads_per_proc, 0, "Limit on threads per proc"); 67 68static int max_threads_hits; 69SYSCTL_INT(_kern_threads, OID_AUTO, max_threads_hits, CTLFLAG_RD, 70 &max_threads_hits, 0, "kern.threads.max_threads_per_proc hit count"); 71 72#ifdef COMPAT_FREEBSD32 73 74static inline int 75suword_lwpid(void *addr, lwpid_t lwpid) 76{ 77 int error; 78 79 if (SV_CURPROC_FLAG(SV_LP64)) 80 error = suword(addr, lwpid); 81 else 82 error = suword32(addr, lwpid); 83 return (error); 84} 85 86#else 87#define suword_lwpid suword 88#endif 89 90/* 91 * System call interface. 92 */ 93 94struct thr_create_initthr_args { 95 ucontext_t ctx; 96 long *tid; 97}; 98 99static int 100thr_create_initthr(struct thread *td, void *thunk) 101{ 102 struct thr_create_initthr_args *args; 103 104 /* Copy out the child tid. */ 105 args = thunk; 106 if (args->tid != NULL && suword_lwpid(args->tid, td->td_tid)) 107 return (EFAULT); 108 109 return (set_mcontext(td, &args->ctx.uc_mcontext)); 110} 111 112int 113sys_thr_create(struct thread *td, struct thr_create_args *uap) 114 /* ucontext_t *ctx, long *id, int flags */ 115{ 116 struct thr_create_initthr_args args; 117 int error; 118 119 if ((error = copyin(uap->ctx, &args.ctx, sizeof(args.ctx)))) 120 return (error); 121 args.tid = uap->id; 122 return (thread_create(td, NULL, thr_create_initthr, &args)); 123} 124 125int 126sys_thr_new(struct thread *td, struct thr_new_args *uap) 127 /* struct thr_param * */ 128{ 129 struct thr_param param; 130 int error; 131 132 if (uap->param_size < 0 || uap->param_size > sizeof(param)) 133 return (EINVAL); 134 bzero(¶m, sizeof(param)); 135 if ((error = copyin(uap->param, ¶m, uap->param_size))) 136 return (error); 137 return (kern_thr_new(td, ¶m)); 138} 139 140static int 141thr_new_initthr(struct thread *td, void *thunk) 142{ 143 stack_t stack; 144 struct thr_param *param; 145 146 /* 147 * Here we copy out tid to two places, one for child and one 148 * for parent, because pthread can create a detached thread, 149 * if parent wants to safely access child tid, it has to provide 150 * its storage, because child thread may exit quickly and 151 * memory is freed before parent thread can access it. 152 */ 153 param = thunk; 154 if ((param->child_tid != NULL && 155 suword_lwpid(param->child_tid, td->td_tid)) || 156 (param->parent_tid != NULL && 157 suword_lwpid(param->parent_tid, td->td_tid))) 158 return (EFAULT); 159 160 /* Set up our machine context. */ 161 stack.ss_sp = param->stack_base; 162 stack.ss_size = param->stack_size; 163 /* Set upcall address to user thread entry function. */ 164 cpu_set_upcall_kse(td, param->start_func, param->arg, &stack); 165 /* Setup user TLS address and TLS pointer register. */ 166 return (cpu_set_user_tls(td, param->tls_base)); 167} 168 169int 170kern_thr_new(struct thread *td, struct thr_param *param) 171{ 172 struct rtprio rtp, *rtpp; 173 int error; 174 175 rtpp = NULL; 176 if (param->rtp != 0) { 177 error = copyin(param->rtp, &rtp, sizeof(struct rtprio)); 178 if (error) 179 return (error); 180 rtpp = &rtp; 181 } 182 return (thread_create(td, rtpp, thr_new_initthr, param)); 183} 184 185int 186thread_create(struct thread *td, struct rtprio *rtp, 187 int (*initialize_thread)(struct thread *, void *), void *thunk) 188{ 189 struct thread *newtd; 190 struct proc *p; 191 int error; 192 193 p = td->td_proc; 194 195 if (rtp != NULL) { 196 switch(rtp->type) { 197 case RTP_PRIO_REALTIME: 198 case RTP_PRIO_FIFO: 199 /* Only root can set scheduler policy */ 200 if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0) 201 return (EPERM); 202 if (rtp->prio > RTP_PRIO_MAX) 203 return (EINVAL); 204 break; 205 case RTP_PRIO_NORMAL: 206 rtp->prio = 0; 207 break; 208 default: 209 return (EINVAL); 210 } 211 } 212 213#ifdef RACCT 214 PROC_LOCK(td->td_proc); 215 error = racct_add(p, RACCT_NTHR, 1); 216 PROC_UNLOCK(td->td_proc); 217 if (error != 0) 218 return (EPROCLIM); 219#endif 220 221 /* Initialize our td */ 222 error = kern_thr_alloc(p, 0, &newtd); 223 if (error) 224 goto fail; 225 226 cpu_set_upcall(newtd, td); 227 228 bzero(&newtd->td_startzero, 229 __rangeof(struct thread, td_startzero, td_endzero)); 230 newtd->td_su = NULL; 231 bcopy(&td->td_startcopy, &newtd->td_startcopy, 232 __rangeof(struct thread, td_startcopy, td_endcopy)); 233 newtd->td_proc = td->td_proc; 234 newtd->td_ucred = crhold(td->td_ucred); 235 newtd->td_dbg_sc_code = td->td_dbg_sc_code; 236 newtd->td_dbg_sc_narg = td->td_dbg_sc_narg; 237 238 error = initialize_thread(newtd, thunk); 239 if (error != 0) { 240 thread_free(newtd); 241 crfree(td->td_ucred); 242 goto fail; 243 } 244 245 PROC_LOCK(td->td_proc); 246 td->td_proc->p_flag |= P_HADTHREADS; 247 thread_link(newtd, p); 248 bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name)); 249 thread_lock(td); 250 /* let the scheduler know about these things. */ 251 sched_fork_thread(td, newtd); 252 thread_unlock(td); 253 if (P_SHOULDSTOP(p)) 254 newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK; 255 if (p->p_flag2 & P2_LWP_EVENTS) 256 newtd->td_dbgflags |= TDB_BORN; 257 PROC_UNLOCK(p); 258 259 tidhash_add(newtd); 260 261 thread_lock(newtd); 262 if (rtp != NULL) { 263 if (!(td->td_pri_class == PRI_TIMESHARE && 264 rtp->type == RTP_PRIO_NORMAL)) { 265 rtp_to_pri(rtp, newtd); 266 sched_prio(newtd, newtd->td_user_pri); 267 } /* ignore timesharing class */ 268 } 269 TD_SET_CAN_RUN(newtd); 270 sched_add(newtd, SRQ_BORING); 271 thread_unlock(newtd); 272 273 return (0); 274 275fail: 276#ifdef RACCT 277 if (racct_enable) { 278 PROC_LOCK(p); 279 racct_sub(p, RACCT_NTHR, 1); 280 PROC_UNLOCK(p); 281 } 282#endif 283 return (error); 284} 285 286int 287sys_thr_self(struct thread *td, struct thr_self_args *uap) 288 /* long *id */ 289{ 290 int error; 291 292 error = suword_lwpid(uap->id, (unsigned)td->td_tid); 293 if (error == -1) 294 return (EFAULT); 295 return (0); 296} 297 298int 299sys_thr_exit(struct thread *td, struct thr_exit_args *uap) 300 /* long *state */ 301{ 302 303 /* Signal userland that it can free the stack. */ 304 if ((void *)uap->state != NULL) { 305 suword_lwpid(uap->state, 1); 306 kern_umtx_wake(td, uap->state, INT_MAX, 0); 307 } 308 309 return (kern_thr_exit(td)); 310} 311 312int 313kern_thr_exit(struct thread *td) 314{ 315 struct proc *p; 316 317 p = td->td_proc; 318 319 /* 320 * If all of the threads in a process call this routine to 321 * exit (e.g. all threads call pthread_exit()), exactly one 322 * thread should return to the caller to terminate the process 323 * instead of the thread. 324 * 325 * Checking p_numthreads alone is not sufficient since threads 326 * might be committed to terminating while the PROC_LOCK is 327 * dropped in either ptracestop() or while removing this thread 328 * from the tidhash. Instead, the p_pendingexits field holds 329 * the count of threads in either of those states and a thread 330 * is considered the "last" thread if all of the other threads 331 * in a process are already terminating. 332 */ 333 PROC_LOCK(p); 334 if (p->p_numthreads == p->p_pendingexits + 1) { 335 /* 336 * Ignore attempts to shut down last thread in the 337 * proc. This will actually call _exit(2) in the 338 * usermode trampoline when it returns. 339 */ 340 PROC_UNLOCK(p); 341 return (0); 342 } 343 344 p->p_pendingexits++; 345 td->td_dbgflags |= TDB_EXIT; 346 if (p->p_flag & P_TRACED && p->p_flag2 & P2_LWP_EVENTS) 347 ptracestop(td, SIGTRAP); 348 PROC_UNLOCK(p); 349 tidhash_remove(td); 350 PROC_LOCK(p); 351 p->p_pendingexits--; 352 353 /* 354 * The check above should prevent all other threads from this 355 * process from exiting while the PROC_LOCK is dropped, so 356 * there must be at least one other thread other than the 357 * current thread. 358 */ 359 KASSERT(p->p_numthreads > 1, ("too few threads")); 360 racct_sub(p, RACCT_NTHR, 1); 361 tdsigcleanup(td); 362 umtx_thread_exit(td); 363 PROC_SLOCK(p); 364 thread_stopped(p); 365 thread_exit(); 366 /* NOTREACHED */ 367} 368 369int 370sys_thr_kill(struct thread *td, struct thr_kill_args *uap) 371 /* long id, int sig */ 372{ 373 ksiginfo_t ksi; 374 struct thread *ttd; 375 struct proc *p; 376 int error; 377 378 p = td->td_proc; 379 ksiginfo_init(&ksi); 380 ksi.ksi_signo = uap->sig; 381 ksi.ksi_code = SI_LWP; 382 ksi.ksi_pid = p->p_pid; 383 ksi.ksi_uid = td->td_ucred->cr_ruid; 384 if (uap->id == -1) { 385 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 386 error = EINVAL; 387 } else { 388 error = ESRCH; 389 PROC_LOCK(p); 390 FOREACH_THREAD_IN_PROC(p, ttd) { 391 if (ttd != td) { 392 error = 0; 393 if (uap->sig == 0) 394 break; 395 tdksignal(ttd, uap->sig, &ksi); 396 } 397 } 398 PROC_UNLOCK(p); 399 } 400 } else { 401 error = 0; 402 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 403 if (ttd == NULL) 404 return (ESRCH); 405 if (uap->sig == 0) 406 ; 407 else if (!_SIG_VALID(uap->sig)) 408 error = EINVAL; 409 else 410 tdksignal(ttd, uap->sig, &ksi); 411 PROC_UNLOCK(ttd->td_proc); 412 } 413 return (error); 414} 415 416int 417sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap) 418 /* pid_t pid, long id, int sig */ 419{ 420 ksiginfo_t ksi; 421 struct thread *ttd; 422 struct proc *p; 423 int error; 424 425 AUDIT_ARG_SIGNUM(uap->sig); 426 427 ksiginfo_init(&ksi); 428 ksi.ksi_signo = uap->sig; 429 ksi.ksi_code = SI_LWP; 430 ksi.ksi_pid = td->td_proc->p_pid; 431 ksi.ksi_uid = td->td_ucred->cr_ruid; 432 if (uap->id == -1) { 433 if ((p = pfind(uap->pid)) == NULL) 434 return (ESRCH); 435 AUDIT_ARG_PROCESS(p); 436 error = p_cansignal(td, p, uap->sig); 437 if (error) { 438 PROC_UNLOCK(p); 439 return (error); 440 } 441 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 442 error = EINVAL; 443 } else { 444 error = ESRCH; 445 FOREACH_THREAD_IN_PROC(p, ttd) { 446 if (ttd != td) { 447 error = 0; 448 if (uap->sig == 0) 449 break; 450 tdksignal(ttd, uap->sig, &ksi); 451 } 452 } 453 } 454 PROC_UNLOCK(p); 455 } else { 456 ttd = tdfind((lwpid_t)uap->id, uap->pid); 457 if (ttd == NULL) 458 return (ESRCH); 459 p = ttd->td_proc; 460 AUDIT_ARG_PROCESS(p); 461 error = p_cansignal(td, p, uap->sig); 462 if (uap->sig == 0) 463 ; 464 else if (!_SIG_VALID(uap->sig)) 465 error = EINVAL; 466 else 467 tdksignal(ttd, uap->sig, &ksi); 468 PROC_UNLOCK(p); 469 } 470 return (error); 471} 472 473int 474sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap) 475 /* const struct timespec *timeout */ 476{ 477 struct timespec ts, *tsp; 478 int error; 479 480 tsp = NULL; 481 if (uap->timeout != NULL) { 482 error = umtx_copyin_timeout(uap->timeout, &ts); 483 if (error != 0) 484 return (error); 485 tsp = &ts; 486 } 487 488 return (kern_thr_suspend(td, tsp)); 489} 490 491int 492kern_thr_suspend(struct thread *td, struct timespec *tsp) 493{ 494 struct proc *p = td->td_proc; 495 struct timeval tv; 496 int error = 0; 497 int timo = 0; 498 499 if (td->td_pflags & TDP_WAKEUP) { 500 td->td_pflags &= ~TDP_WAKEUP; 501 return (0); 502 } 503 504 if (tsp != NULL) { 505 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 506 error = EWOULDBLOCK; 507 else { 508 TIMESPEC_TO_TIMEVAL(&tv, tsp); 509 timo = tvtohz(&tv); 510 } 511 } 512 513 PROC_LOCK(p); 514 if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0) 515 error = msleep((void *)td, &p->p_mtx, 516 PCATCH, "lthr", timo); 517 518 if (td->td_flags & TDF_THRWAKEUP) { 519 thread_lock(td); 520 td->td_flags &= ~TDF_THRWAKEUP; 521 thread_unlock(td); 522 PROC_UNLOCK(p); 523 return (0); 524 } 525 PROC_UNLOCK(p); 526 if (error == EWOULDBLOCK) 527 error = ETIMEDOUT; 528 else if (error == ERESTART) { 529 if (timo != 0) 530 error = EINTR; 531 } 532 return (error); 533} 534 535int 536sys_thr_wake(struct thread *td, struct thr_wake_args *uap) 537 /* long id */ 538{ 539 struct proc *p; 540 struct thread *ttd; 541 542 if (uap->id == td->td_tid) { 543 td->td_pflags |= TDP_WAKEUP; 544 return (0); 545 } 546 547 p = td->td_proc; 548 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 549 if (ttd == NULL) 550 return (ESRCH); 551 thread_lock(ttd); 552 ttd->td_flags |= TDF_THRWAKEUP; 553 thread_unlock(ttd); 554 wakeup((void *)ttd); 555 PROC_UNLOCK(p); 556 return (0); 557} 558 559int 560sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap) 561{ 562 struct proc *p; 563 char name[MAXCOMLEN + 1]; 564 struct thread *ttd; 565 int error; 566 567 error = 0; 568 name[0] = '\0'; 569 if (uap->name != NULL) { 570 error = copyinstr(uap->name, name, sizeof(name), 571 NULL); 572 if (error) 573 return (error); 574 } 575 p = td->td_proc; 576 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 577 if (ttd == NULL) 578 return (ESRCH); 579 strcpy(ttd->td_name, name); 580#ifdef KTR 581 sched_clear_tdname(ttd); 582#endif 583 PROC_UNLOCK(p); 584 return (error); 585} 586 587int 588kern_thr_alloc(struct proc *p, int pages, struct thread **ntd) 589{ 590 591 /* Have race condition but it is cheap. */ 592 if (p->p_numthreads >= max_threads_per_proc) { 593 ++max_threads_hits; 594 return (EPROCLIM); 595 } 596 597 *ntd = thread_alloc(pages); 598 if (*ntd == NULL) 599 return (ENOMEM); 600 601 return (0); 602} 603