sys/kern/kern_thr.c

   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("$FreeBSD$");
  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/resourcevar.h>
  40 #include <sys/sched.h>
  41 #include <sys/sysctl.h>
  42 #include <sys/smp.h>
  43 #include <sys/syscallsubr.h>
  44 #include <sys/sysent.h>
  45 #include <sys/systm.h>
  46 #include <sys/sysproto.h>
  47 #include <sys/signalvar.h>
  48 #include <sys/ucontext.h>
  49 #include <sys/thr.h>
  50 #include <sys/rtprio.h>
  51 #include <sys/umtx.h>
  52 #include <sys/limits.h>
  53
  54 #include <machine/frame.h>
  55
  56 #include <security/audit/audit.h>
  57
  58 #ifdef COMPAT_IA32
  59
  60 extern struct sysentvec ia32_freebsd_sysvec;
  61
  62 static inline int
  63 suword_lwpid(void *addr, lwpid_t lwpid)
  64 {
  65         int error;
  66
  67         if (curproc->p_sysent != &ia32_freebsd_sysvec)
  68                 error = suword(addr, lwpid);
  69         else
  70                 error = suword32(addr, lwpid);
  71         return (error);
  72 }
  73
  74 #else
  75 #define suword_lwpid    suword
  76 #endif
  77
  78 extern int max_threads_per_proc;
  79
  80 static int create_thread(struct thread *td, mcontext_t *ctx,
  81                          void (*start_func)(void *), void *arg,
  82                          char *stack_base, size_t stack_size,
  83                          char *tls_base,
  84                          long *child_tid, long *parent_tid,
  85                          int flags, struct rtprio *rtp);
  86
  87 /*
  88  * System call interface.
  89  */
  90 int
  91 thr_create(struct thread *td, struct thr_create_args *uap)
  92     /* ucontext_t *ctx, long *id, int flags */
  93 {
  94         ucontext_t ctx;
  95         int error;
  96
  97         if ((error = copyin(uap->ctx, &ctx, sizeof(ctx))))
  98                 return (error);
  99
 100         error = create_thread(td, &ctx.uc_mcontext, NULL, NULL,
 101                 NULL, 0, NULL, uap->id, NULL, uap->flags, NULL);
 102         return (error);
 103 }
 104
 105 int
 106 thr_new(struct thread *td, struct thr_new_args *uap)
 107     /* struct thr_param * */
 108 {
 109         struct thr_param param;
 110         int error;
 111
 112         if (uap->param_size < 0 || uap->param_size > sizeof(param))
 113                 return (EINVAL);
 114         bzero(&param, sizeof(param));
 115         if ((error = copyin(uap->param, &param, uap->param_size)))
 116                 return (error);
 117         return (kern_thr_new(td, &param));
 118 }
 119
 120 int
 121 kern_thr_new(struct thread *td, struct thr_param *param)
 122 {
 123         struct rtprio rtp, *rtpp;
 124         int error;
 125
 126         rtpp = NULL;
 127         if (param->rtp != 0) {
 128                 error = copyin(param->rtp, &rtp, sizeof(struct rtprio));
 129                 rtpp = &rtp;
 130         }
 131         error = create_thread(td, NULL, param->start_func, param->arg,
 132                 param->stack_base, param->stack_size, param->tls_base,
 133                 param->child_tid, param->parent_tid, param->flags,
 134                 rtpp);
 135         return (error);
 136 }
 137
 138 static int
 139 create_thread(struct thread *td, mcontext_t *ctx,
 140             void (*start_func)(void *), void *arg,
 141             char *stack_base, size_t stack_size,
 142             char *tls_base,
 143             long *child_tid, long *parent_tid,
 144             int flags, struct rtprio *rtp)
 145 {
 146         stack_t stack;
 147         struct thread *newtd;
 148         struct proc *p;
 149         int error;
 150
 151         error = 0;
 152         p = td->td_proc;
 153
 154         /* Have race condition but it is cheap. */
 155         if (p->p_numthreads >= max_threads_per_proc)
 156                 return (EPROCLIM);
 157
 158         if (rtp != NULL) {
 159                 switch(rtp->type) {
 160                 case RTP_PRIO_REALTIME:
 161                 case RTP_PRIO_FIFO:
 162                         /* Only root can set scheduler policy */
 163                         if (priv_check(td, PRIV_SCHED_SETPOLICY) != 0)
 164                                 return (EPERM);
 165                         if (rtp->prio > RTP_PRIO_MAX)
 166                                 return (EINVAL);
 167                         break;
 168                 case RTP_PRIO_NORMAL:
 169                         rtp->prio = 0;
 170                         break;
 171                 default:
 172                         return (EINVAL);
 173                 }
 174         }
 175
 176         /* Initialize our td */
 177         newtd = thread_alloc();
 178         if (newtd == NULL)
 179                 return (ENOMEM);
 180
 181         /*
 182          * Try the copyout as soon as we allocate the td so we don't
 183          * have to tear things down in a failure case below.
 184          * Here we copy out tid to two places, one for child and one
 185          * for parent, because pthread can create a detached thread,
 186          * if parent wants to safely access child tid, it has to provide
 187          * its storage, because child thread may exit quickly and
 188          * memory is freed before parent thread can access it.
 189          */
 190         if ((child_tid != NULL &&
 191             suword_lwpid(child_tid, newtd->td_tid)) ||
 192             (parent_tid != NULL &&
 193             suword_lwpid(parent_tid, newtd->td_tid))) {
 194                 thread_free(newtd);
 195                 return (EFAULT);
 196         }
 197
 198         bzero(&newtd->td_startzero,
 199             __rangeof(struct thread, td_startzero, td_endzero));
 200         bcopy(&td->td_startcopy, &newtd->td_startcopy,
 201             __rangeof(struct thread, td_startcopy, td_endcopy));
 202         newtd->td_proc = td->td_proc;
 203         newtd->td_ucred = crhold(td->td_ucred);
 204
 205         cpu_set_upcall(newtd, td);
 206
 207         if (ctx != NULL) { /* old way to set user context */
 208                 error = set_mcontext(newtd, ctx);
 209                 if (error != 0) {
 210                         thread_free(newtd);
 211                         crfree(td->td_ucred);
 212                         return (error);
 213                 }
 214         } else {
 215                 /* Set up our machine context. */
 216                 stack.ss_sp = stack_base;
 217                 stack.ss_size = stack_size;
 218                 /* Set upcall address to user thread entry function. */
 219                 cpu_set_upcall_kse(newtd, start_func, arg, &stack);
 220                 /* Setup user TLS address and TLS pointer register. */
 221                 error = cpu_set_user_tls(newtd, tls_base);
 222                 if (error != 0) {
 223                         thread_free(newtd);
 224                         crfree(td->td_ucred);
 225                         return (error);
 226                 }
 227         }
 228
 229         PROC_LOCK(td->td_proc);
 230         td->td_proc->p_flag |= P_HADTHREADS;
 231         newtd->td_sigmask = td->td_sigmask;
 232         thread_link(newtd, p);
 233         bcopy(p->p_comm, newtd->td_name, sizeof(newtd->td_name));
 234         thread_lock(td);
 235         /* let the scheduler know about these things. */
 236         sched_fork_thread(td, newtd);
 237         thread_unlock(td);
 238         if (P_SHOULDSTOP(p))
 239                 newtd->td_flags |= TDF_ASTPENDING | TDF_NEEDSUSPCHK;
 240         PROC_UNLOCK(p);
 241         thread_lock(newtd);
 242         if (rtp != NULL) {
 243                 if (!(td->td_pri_class == PRI_TIMESHARE &&
 244                       rtp->type == RTP_PRIO_NORMAL)) {
 245                         rtp_to_pri(rtp, newtd);
 246                         sched_prio(newtd, newtd->td_user_pri);
 247                 } /* ignore timesharing class */
 248         }
 249         TD_SET_CAN_RUN(newtd);
 250         sched_add(newtd, SRQ_BORING);
 251         thread_unlock(newtd);
 252
 253         return (error);
 254 }
 255
 256 int
 257 thr_self(struct thread *td, struct thr_self_args *uap)
 258     /* long *id */
 259 {
 260         int error;
 261
 262         error = suword_lwpid(uap->id, (unsigned)td->td_tid);
 263         if (error == -1)
 264                 return (EFAULT);
 265         return (0);
 266 }
 267
 268 int
 269 thr_exit(struct thread *td, struct thr_exit_args *uap)
 270     /* long *state */
 271 {
 272         struct proc *p;
 273
 274         p = td->td_proc;
 275
 276         /* Signal userland that it can free the stack. */
 277         if ((void *)uap->state != NULL) {
 278                 suword_lwpid(uap->state, 1);
 279                 kern_umtx_wake(td, uap->state, INT_MAX, 0);
 280         }
 281
 282         PROC_LOCK(p);
 283         sigqueue_flush(&td->td_sigqueue);
 284         PROC_SLOCK(p);
 285
 286         /*
 287          * Shutting down last thread in the proc.  This will actually
 288          * call exit() in the trampoline when it returns.
 289          */
 290         if (p->p_numthreads != 1) {
 291                 thread_stopped(p);
 292                 thread_exit();
 293                 /* NOTREACHED */
 294         }
 295         PROC_SUNLOCK(p);
 296         PROC_UNLOCK(p);
 297         return (0);
 298 }
 299
 300 int
 301 thr_kill(struct thread *td, struct thr_kill_args *uap)
 302     /* long id, int sig */
 303 {
 304         struct thread *ttd;
 305         struct proc *p;
 306         int error;
 307
 308         p = td->td_proc;
 309         error = 0;
 310         PROC_LOCK(p);
 311         if (uap->id == -1) {
 312                 if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
 313                         error = EINVAL;
 314                 } else {
 315                         error = ESRCH;
 316                         FOREACH_THREAD_IN_PROC(p, ttd) {
 317                                 if (ttd != td) {
 318                                         error = 0;
 319                                         if (uap->sig == 0)
 320                                                 break;
 321                                         tdsignal(p, ttd, uap->sig, NULL);
 322                                 }
 323                         }
 324                 }
 325         } else {
 326                 if (uap->id != td->td_tid)
 327                         ttd = thread_find(p, uap->id);
 328                 else
 329                         ttd = td;
 330                 if (ttd == NULL)
 331                         error = ESRCH;
 332                 else if (uap->sig == 0)
 333                         ;
 334                 else if (!_SIG_VALID(uap->sig))
 335                         error = EINVAL;
 336                 else
 337                         tdsignal(p, ttd, uap->sig, NULL);
 338         }
 339         PROC_UNLOCK(p);
 340         return (error);
 341 }
 342
 343 int
 344 thr_kill2(struct thread *td, struct thr_kill2_args *uap)
 345     /* pid_t pid, long id, int sig */
 346 {
 347         struct thread *ttd;
 348         struct proc *p;
 349         int error;
 350
 351         AUDIT_ARG(signum, uap->sig);
 352
 353         if (uap->pid == td->td_proc->p_pid) {
 354                 p = td->td_proc;
 355                 PROC_LOCK(p);
 356         } else if ((p = pfind(uap->pid)) == NULL) {
 357                 return (ESRCH);
 358         }
 359         AUDIT_ARG(process, p);
 360
 361         error = p_cansignal(td, p, uap->sig);
 362         if (error == 0) {
 363                 if (uap->id == -1) {
 364                         if (uap->sig != 0 && !_SIG_VALID(uap->sig)) {
 365                                 error = EINVAL;
 366                         } else {
 367                                 error = ESRCH;
 368                                 FOREACH_THREAD_IN_PROC(p, ttd) {
 369                                         if (ttd != td) {
 370                                                 error = 0;
 371                                                 if (uap->sig == 0)
 372                                                         break;
 373                                                 tdsignal(p, ttd, uap->sig, NULL);
 374                                         }
 375                                 }
 376                         }
 377                 } else {
 378                         if (uap->id != td->td_tid)
 379                                 ttd = thread_find(p, uap->id);
 380                         else
 381                                 ttd = td;
 382                         if (ttd == NULL)
 383                                 error = ESRCH;
 384                         else if (uap->sig == 0)
 385                                 ;
 386                         else if (!_SIG_VALID(uap->sig))
 387                                 error = EINVAL;
 388                         else
 389                                 tdsignal(p, ttd, uap->sig, NULL);
 390                 }
 391         }
 392         PROC_UNLOCK(p);
 393         return (error);
 394 }
 395
 396 int
 397 thr_suspend(struct thread *td, struct thr_suspend_args *uap)
 398         /* const struct timespec *timeout */
 399 {
 400         struct timespec ts, *tsp;
 401         int error;
 402
 403         error = 0;
 404         tsp = NULL;
 405         if (uap->timeout != NULL) {
 406                 error = copyin((const void *)uap->timeout, (void *)&ts,
 407                     sizeof(struct timespec));
 408                 if (error != 0)
 409                         return (error);
 410                 tsp = &ts;
 411         }
 412
 413         return (kern_thr_suspend(td, tsp));
 414 }
 415
 416 int
 417 kern_thr_suspend(struct thread *td, struct timespec *tsp)
 418 {
 419         struct timeval tv;
 420         int error = 0, hz = 0;
 421
 422         if (tsp != NULL) {
 423                 if (tsp->tv_nsec < 0 || tsp->tv_nsec > 1000000000)
 424                         return (EINVAL);
 425                 if (tsp->tv_sec == 0 && tsp->tv_nsec == 0)
 426                         return (ETIMEDOUT);
 427                 TIMESPEC_TO_TIMEVAL(&tv, tsp);
 428                 hz = tvtohz(&tv);
 429         }
 430
 431         if (td->td_pflags & TDP_WAKEUP) {
 432                 td->td_pflags &= ~TDP_WAKEUP;
 433                 return (0);
 434         }
 435
 436         PROC_LOCK(td->td_proc);
 437         if ((td->td_flags & TDF_THRWAKEUP) == 0)
 438                 error = msleep((void *)td, &td->td_proc->p_mtx, PCATCH, "lthr",
 439                     hz);
 440         if (td->td_flags & TDF_THRWAKEUP) {
 441                 thread_lock(td);
 442                 td->td_flags &= ~TDF_THRWAKEUP;
 443                 thread_unlock(td);
 444                 PROC_UNLOCK(td->td_proc);
 445                 return (0);
 446         }
 447         PROC_UNLOCK(td->td_proc);
 448         if (error == EWOULDBLOCK)
 449                 error = ETIMEDOUT;
 450         else if (error == ERESTART) {
 451                 if (hz != 0)
 452                         error = EINTR;
 453         }
 454         return (error);
 455 }
 456
 457 int
 458 thr_wake(struct thread *td, struct thr_wake_args *uap)
 459         /* long id */
 460 {
 461         struct proc *p;
 462         struct thread *ttd;
 463
 464         if (uap->id == td->td_tid) {
 465                 td->td_pflags |= TDP_WAKEUP;
 466                 return (0);
 467         }
 468
 469         p = td->td_proc;
 470         PROC_LOCK(p);
 471         ttd = thread_find(p, uap->id);
 472         if (ttd == NULL) {
 473                 PROC_UNLOCK(p);
 474                 return (ESRCH);
 475         }
 476         thread_lock(ttd);
 477         ttd->td_flags |= TDF_THRWAKEUP;
 478         thread_unlock(ttd);
 479         wakeup((void *)ttd);
 480         PROC_UNLOCK(p);
 481         return (0);
 482 }
 483
 484 int
 485 thr_set_name(struct thread *td, struct thr_set_name_args *uap)
 486 {
 487         struct proc *p = td->td_proc;
 488         char name[MAXCOMLEN + 1];
 489         struct thread *ttd;
 490         int error;
 491
 492         error = 0;
 493         name[0] = '\0';
 494         if (uap->name != NULL) {
 495                 error = copyinstr(uap->name, name, sizeof(name),
 496                         NULL);
 497                 if (error)
 498                         return (error);
 499         }
 500         PROC_LOCK(p);
 501         if (uap->id == td->td_tid)
 502                 ttd = td;
 503         else
 504                 ttd = thread_find(p, uap->id);
 505         if (ttd != NULL)
 506                 strcpy(ttd->td_name, name);
 507         else
 508                 error = ESRCH;
 509         PROC_UNLOCK(p);
 510         return (error);
 511 }