kern_thr.c revision 296373
11984Scsgr/*- 250476Speter * Copyright (c) 2003, Jeffrey Roberson <jeff@freebsd.org> 31984Scsgr * All rights reserved. 41984Scsgr * 5156837Sru * Redistribution and use in source and binary forms, with or without 6156837Sru * modification, are permitted provided that the following conditions 7156813Sru * are met: 8156813Sru * 1. Redistributions of source code must retain the above copyright 9195767Skensmith * notice unmodified, this list of conditions, and the following 1070419Speter * disclaimer. 1129146Speter * 2. Redistributions in binary form must reproduce the above copyright 1251462Smarkm * notice, this list of conditions and the following disclaimer in the 13115720Smarkm * documentation and/or other materials provided with the distribution. 14115720Smarkm * 15115720Smarkm * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 1674870Sru * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 1764918Sgreen * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 1874607Speter * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19115720Smarkm * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20115720Smarkm * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21156813Sru * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 2270419Speter * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 2374106Smarkm * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 2474106Smarkm * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 2570419Speter */ 26115720Smarkm 2770419Speter#include <sys/cdefs.h> 2870419Speter__FBSDID("$FreeBSD: releng/10.3/sys/kern/kern_thr.c 293481 2016-01-09 14:38:29Z dchagin $"); 2970419Speter 30115720Smarkm#include "opt_compat.h" 31115720Smarkm#include "opt_posix.h" 32115720Smarkm#include <sys/param.h> 3370419Speter#include <sys/kernel.h> 3470419Speter#include <sys/lock.h> 35115720Smarkm#include <sys/mutex.h> 36201381Sed#include <sys/priv.h> 37201381Sed#include <sys/proc.h> 38136910Sru#include <sys/posix4.h> 3943092Smarkm#include <sys/racct.h> 401984Scsgr#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 PROC_UNLOCK(p); 256 257 tidhash_add(newtd); 258 259 thread_lock(newtd); 260 if (rtp != NULL) { 261 if (!(td->td_pri_class == PRI_TIMESHARE && 262 rtp->type == RTP_PRIO_NORMAL)) { 263 rtp_to_pri(rtp, newtd); 264 sched_prio(newtd, newtd->td_user_pri); 265 } /* ignore timesharing class */ 266 } 267 TD_SET_CAN_RUN(newtd); 268 sched_add(newtd, SRQ_BORING); 269 thread_unlock(newtd); 270 271 return (0); 272 273fail: 274#ifdef RACCT 275 if (racct_enable) { 276 PROC_LOCK(p); 277 racct_sub(p, RACCT_NTHR, 1); 278 PROC_UNLOCK(p); 279 } 280#endif 281 return (error); 282} 283 284int 285sys_thr_self(struct thread *td, struct thr_self_args *uap) 286 /* long *id */ 287{ 288 int error; 289 290 error = suword_lwpid(uap->id, (unsigned)td->td_tid); 291 if (error == -1) 292 return (EFAULT); 293 return (0); 294} 295 296int 297sys_thr_exit(struct thread *td, struct thr_exit_args *uap) 298 /* long *state */ 299{ 300 301 /* Signal userland that it can free the stack. */ 302 if ((void *)uap->state != NULL) { 303 suword_lwpid(uap->state, 1); 304 kern_umtx_wake(td, uap->state, INT_MAX, 0); 305 } 306 307 return (kern_thr_exit(td)); 308} 309 310int 311kern_thr_exit(struct thread *td) 312{ 313 struct proc *p; 314 315 p = td->td_proc; 316 317 rw_wlock(&tidhash_lock); 318 PROC_LOCK(p); 319 320 if (p->p_numthreads != 1) { 321 racct_sub(p, RACCT_NTHR, 1); 322 LIST_REMOVE(td, td_hash); 323 rw_wunlock(&tidhash_lock); 324 tdsigcleanup(td); 325 umtx_thread_exit(td); 326 PROC_SLOCK(p); 327 thread_stopped(p); 328 thread_exit(); 329 /* NOTREACHED */ 330 } 331 332 /* 333 * Ignore attempts to shut down last thread in the proc. This 334 * will actually call _exit(2) in the usermode trampoline when 335 * it returns. 336 */ 337 PROC_UNLOCK(p); 338 rw_wunlock(&tidhash_lock); 339 return (0); 340} 341 342int 343sys_thr_kill(struct thread *td, struct thr_kill_args *uap) 344 /* long id, int sig */ 345{ 346 ksiginfo_t ksi; 347 struct thread *ttd; 348 struct proc *p; 349 int error; 350 351 p = td->td_proc; 352 ksiginfo_init(&ksi); 353 ksi.ksi_signo = uap->sig; 354 ksi.ksi_code = SI_LWP; 355 ksi.ksi_pid = p->p_pid; 356 ksi.ksi_uid = td->td_ucred->cr_ruid; 357 if (uap->id == -1) { 358 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 359 error = EINVAL; 360 } else { 361 error = ESRCH; 362 PROC_LOCK(p); 363 FOREACH_THREAD_IN_PROC(p, ttd) { 364 if (ttd != td) { 365 error = 0; 366 if (uap->sig == 0) 367 break; 368 tdksignal(ttd, uap->sig, &ksi); 369 } 370 } 371 PROC_UNLOCK(p); 372 } 373 } else { 374 error = 0; 375 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 376 if (ttd == NULL) 377 return (ESRCH); 378 if (uap->sig == 0) 379 ; 380 else if (!_SIG_VALID(uap->sig)) 381 error = EINVAL; 382 else 383 tdksignal(ttd, uap->sig, &ksi); 384 PROC_UNLOCK(ttd->td_proc); 385 } 386 return (error); 387} 388 389int 390sys_thr_kill2(struct thread *td, struct thr_kill2_args *uap) 391 /* pid_t pid, long id, int sig */ 392{ 393 ksiginfo_t ksi; 394 struct thread *ttd; 395 struct proc *p; 396 int error; 397 398 AUDIT_ARG_SIGNUM(uap->sig); 399 400 ksiginfo_init(&ksi); 401 ksi.ksi_signo = uap->sig; 402 ksi.ksi_code = SI_LWP; 403 ksi.ksi_pid = td->td_proc->p_pid; 404 ksi.ksi_uid = td->td_ucred->cr_ruid; 405 if (uap->id == -1) { 406 if ((p = pfind(uap->pid)) == NULL) 407 return (ESRCH); 408 AUDIT_ARG_PROCESS(p); 409 error = p_cansignal(td, p, uap->sig); 410 if (error) { 411 PROC_UNLOCK(p); 412 return (error); 413 } 414 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) { 415 error = EINVAL; 416 } else { 417 error = ESRCH; 418 FOREACH_THREAD_IN_PROC(p, ttd) { 419 if (ttd != td) { 420 error = 0; 421 if (uap->sig == 0) 422 break; 423 tdksignal(ttd, uap->sig, &ksi); 424 } 425 } 426 } 427 PROC_UNLOCK(p); 428 } else { 429 ttd = tdfind((lwpid_t)uap->id, uap->pid); 430 if (ttd == NULL) 431 return (ESRCH); 432 p = ttd->td_proc; 433 AUDIT_ARG_PROCESS(p); 434 error = p_cansignal(td, p, uap->sig); 435 if (uap->sig == 0) 436 ; 437 else if (!_SIG_VALID(uap->sig)) 438 error = EINVAL; 439 else 440 tdksignal(ttd, uap->sig, &ksi); 441 PROC_UNLOCK(p); 442 } 443 return (error); 444} 445 446int 447sys_thr_suspend(struct thread *td, struct thr_suspend_args *uap) 448 /* const struct timespec *timeout */ 449{ 450 struct timespec ts, *tsp; 451 int error; 452 453 tsp = NULL; 454 if (uap->timeout != NULL) { 455 error = umtx_copyin_timeout(uap->timeout, &ts); 456 if (error != 0) 457 return (error); 458 tsp = &ts; 459 } 460 461 return (kern_thr_suspend(td, tsp)); 462} 463 464int 465kern_thr_suspend(struct thread *td, struct timespec *tsp) 466{ 467 struct proc *p = td->td_proc; 468 struct timeval tv; 469 int error = 0; 470 int timo = 0; 471 472 if (td->td_pflags & TDP_WAKEUP) { 473 td->td_pflags &= ~TDP_WAKEUP; 474 return (0); 475 } 476 477 if (tsp != NULL) { 478 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0) 479 error = EWOULDBLOCK; 480 else { 481 TIMESPEC_TO_TIMEVAL(&tv, tsp); 482 timo = tvtohz(&tv); 483 } 484 } 485 486 PROC_LOCK(p); 487 if (error == 0 && (td->td_flags & TDF_THRWAKEUP) == 0) 488 error = msleep((void *)td, &p->p_mtx, 489 PCATCH, "lthr", timo); 490 491 if (td->td_flags & TDF_THRWAKEUP) { 492 thread_lock(td); 493 td->td_flags &= ~TDF_THRWAKEUP; 494 thread_unlock(td); 495 PROC_UNLOCK(p); 496 return (0); 497 } 498 PROC_UNLOCK(p); 499 if (error == EWOULDBLOCK) 500 error = ETIMEDOUT; 501 else if (error == ERESTART) { 502 if (timo != 0) 503 error = EINTR; 504 } 505 return (error); 506} 507 508int 509sys_thr_wake(struct thread *td, struct thr_wake_args *uap) 510 /* long id */ 511{ 512 struct proc *p; 513 struct thread *ttd; 514 515 if (uap->id == td->td_tid) { 516 td->td_pflags |= TDP_WAKEUP; 517 return (0); 518 } 519 520 p = td->td_proc; 521 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 522 if (ttd == NULL) 523 return (ESRCH); 524 thread_lock(ttd); 525 ttd->td_flags |= TDF_THRWAKEUP; 526 thread_unlock(ttd); 527 wakeup((void *)ttd); 528 PROC_UNLOCK(p); 529 return (0); 530} 531 532int 533sys_thr_set_name(struct thread *td, struct thr_set_name_args *uap) 534{ 535 struct proc *p; 536 char name[MAXCOMLEN + 1]; 537 struct thread *ttd; 538 int error; 539 540 error = 0; 541 name[0] = '\0'; 542 if (uap->name != NULL) { 543 error = copyinstr(uap->name, name, sizeof(name), 544 NULL); 545 if (error) 546 return (error); 547 } 548 p = td->td_proc; 549 ttd = tdfind((lwpid_t)uap->id, p->p_pid); 550 if (ttd == NULL) 551 return (ESRCH); 552 strcpy(ttd->td_name, name); 553#ifdef KTR 554 sched_clear_tdname(ttd); 555#endif 556 PROC_UNLOCK(p); 557 return (error); 558} 559 560int 561kern_thr_alloc(struct proc *p, int pages, struct thread **ntd) 562{ 563 564 /* Have race condition but it is cheap. */ 565 if (p->p_numthreads >= max_threads_per_proc) { 566 ++max_threads_hits; 567 return (EPROCLIM); 568 } 569 570 *ntd = thread_alloc(pages); 571 if (*ntd == NULL) 572 return (ENOMEM); 573 574 return (0); 575} 576