Home | History | Annotate | Line # | Download | only in kern
sys_lwp.c revision 1.68
      1  1.68      maxv /*	$NetBSD: sys_lwp.c,v 1.68 2019/07/01 17:15:43 maxv Exp $	*/
      2   1.2        ad 
      3   1.2        ad /*-
      4  1.36        ad  * Copyright (c) 2001, 2006, 2007, 2008 The NetBSD Foundation, Inc.
      5   1.2        ad  * All rights reserved.
      6   1.2        ad  *
      7   1.2        ad  * This code is derived from software contributed to The NetBSD Foundation
      8   1.2        ad  * by Nathan J. Williams, and Andrew Doran.
      9   1.2        ad  *
     10   1.2        ad  * Redistribution and use in source and binary forms, with or without
     11   1.2        ad  * modification, are permitted provided that the following conditions
     12   1.2        ad  * are met:
     13   1.2        ad  * 1. Redistributions of source code must retain the above copyright
     14   1.2        ad  *    notice, this list of conditions and the following disclaimer.
     15   1.2        ad  * 2. Redistributions in binary form must reproduce the above copyright
     16   1.2        ad  *    notice, this list of conditions and the following disclaimer in the
     17   1.2        ad  *    documentation and/or other materials provided with the distribution.
     18   1.2        ad  *
     19   1.2        ad  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20   1.2        ad  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21   1.2        ad  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22   1.2        ad  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23   1.2        ad  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24   1.2        ad  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25   1.2        ad  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26   1.2        ad  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27   1.2        ad  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28   1.2        ad  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29   1.2        ad  * POSSIBILITY OF SUCH DAMAGE.
     30   1.2        ad  */
     31   1.2        ad 
     32   1.2        ad /*
     33   1.2        ad  * Lightweight process (LWP) system calls.  See kern_lwp.c for a description
     34   1.2        ad  * of LWPs.
     35   1.2        ad  */
     36   1.2        ad 
     37   1.2        ad #include <sys/cdefs.h>
     38  1.68      maxv __KERNEL_RCSID(0, "$NetBSD: sys_lwp.c,v 1.68 2019/07/01 17:15:43 maxv Exp $");
     39   1.2        ad 
     40   1.2        ad #include <sys/param.h>
     41   1.2        ad #include <sys/systm.h>
     42   1.2        ad #include <sys/pool.h>
     43   1.2        ad #include <sys/proc.h>
     44   1.2        ad #include <sys/types.h>
     45   1.2        ad #include <sys/syscallargs.h>
     46   1.2        ad #include <sys/kauth.h>
     47   1.2        ad #include <sys/kmem.h>
     48   1.2        ad #include <sys/sleepq.h>
     49  1.30        ad #include <sys/lwpctl.h>
     50  1.45        ad #include <sys/cpu.h>
     51   1.2        ad 
     52   1.2        ad #include <uvm/uvm_extern.h>
     53   1.2        ad 
     54   1.2        ad #define	LWP_UNPARK_MAX		1024
     55   1.2        ad 
     56  1.47     rmind static syncobj_t lwp_park_sobj = {
     57  1.63     ozaki 	.sobj_flag	= SOBJ_SLEEPQ_LIFO,
     58  1.63     ozaki 	.sobj_unsleep	= sleepq_unsleep,
     59  1.63     ozaki 	.sobj_changepri	= sleepq_changepri,
     60  1.63     ozaki 	.sobj_lendpri	= sleepq_lendpri,
     61  1.63     ozaki 	.sobj_owner	= syncobj_noowner,
     62   1.2        ad };
     63   1.2        ad 
     64  1.47     rmind static sleeptab_t	lwp_park_tab;
     65   1.2        ad 
     66   1.2        ad void
     67   1.2        ad lwp_sys_init(void)
     68   1.2        ad {
     69   1.2        ad 	sleeptab_init(&lwp_park_tab);
     70   1.2        ad }
     71   1.2        ad 
     72  1.64     kamil static void
     73  1.64     kamil mi_startlwp(void *arg)
     74  1.64     kamil {
     75  1.64     kamil 	struct lwp *l = curlwp;
     76  1.64     kamil 	struct proc *p = l->l_proc;
     77  1.64     kamil 
     78  1.65     kamil 	(p->p_emul->e_startlwp)(arg);
     79  1.65     kamil 
     80  1.64     kamil 	/* If the process is traced, report lwp creation to a debugger */
     81  1.66     kamil 	if ((p->p_slflag & (PSL_TRACED|PSL_TRACELWP_CREATE)) ==
     82  1.64     kamil 	    (PSL_TRACED|PSL_TRACELWP_CREATE)) {
     83  1.64     kamil 		/* Paranoid check */
     84  1.64     kamil 		mutex_enter(proc_lock);
     85  1.66     kamil 		if ((p->p_slflag & (PSL_TRACED|PSL_TRACELWP_CREATE)) !=
     86  1.66     kamil 		    (PSL_TRACED|PSL_TRACELWP_CREATE)) {
     87  1.64     kamil 			mutex_exit(proc_lock);
     88  1.65     kamil 			return;
     89  1.64     kamil 		}
     90  1.64     kamil 
     91  1.64     kamil 		mutex_enter(p->p_lock);
     92  1.64     kamil 		p->p_lwp_created = l->l_lid;
     93  1.67     kamil 		eventswitch(TRAP_LWP);
     94  1.64     kamil 	}
     95  1.64     kamil }
     96  1.64     kamil 
     97   1.2        ad int
     98  1.59  christos do_lwp_create(lwp_t *l, void *arg, u_long flags, lwpid_t *new_lwp,
     99  1.59  christos     const sigset_t *sigmask, const stack_t *sigstk)
    100   1.2        ad {
    101   1.2        ad 	struct proc *p = l->l_proc;
    102   1.2        ad 	struct lwp *l2;
    103  1.50     skrll 	struct schedstate_percpu *spc;
    104   1.2        ad 	vaddr_t uaddr;
    105  1.54    martin 	int error;
    106   1.2        ad 
    107   1.2        ad 	/* XXX check against resource limits */
    108   1.2        ad 
    109  1.46     rmind 	uaddr = uvm_uarea_alloc();
    110  1.54    martin 	if (__predict_false(uaddr == 0))
    111   1.2        ad 		return ENOMEM;
    112   1.2        ad 
    113  1.59  christos 	error = lwp_create(l, p, uaddr, flags & LWP_DETACHED, NULL, 0,
    114  1.64     kamil 	    mi_startlwp, arg, &l2, l->l_class, sigmask, &SS_INIT);
    115  1.46     rmind 	if (__predict_false(error)) {
    116  1.46     rmind 		uvm_uarea_free(uaddr);
    117  1.18     rmind 		return error;
    118  1.18     rmind 	}
    119   1.2        ad 
    120  1.54    martin 	*new_lwp = l2->l_lid;
    121  1.21     rmind 
    122   1.2        ad 	/*
    123   1.2        ad 	 * Set the new LWP running, unless the caller has requested that
    124   1.2        ad 	 * it be created in suspended state.  If the process is stopping,
    125   1.2        ad 	 * then the LWP is created stopped.
    126   1.2        ad 	 */
    127  1.39        ad 	mutex_enter(p->p_lock);
    128   1.2        ad 	lwp_lock(l2);
    129  1.50     skrll 	spc = &l2->l_cpu->ci_schedstate;
    130  1.54    martin 	if ((flags & LWP_SUSPENDED) == 0 &&
    131   1.4     pavel 	    (l->l_flag & (LW_WREBOOT | LW_WSUSPEND | LW_WEXIT)) == 0) {
    132  1.50     skrll 	    	if (p->p_stat == SSTOP || (p->p_sflag & PS_STOPPING) != 0) {
    133  1.50     skrll 			KASSERT(l2->l_wchan == NULL);
    134   1.2        ad 	    		l2->l_stat = LSSTOP;
    135  1.51      yamt 			p->p_nrlwps--;
    136  1.50     skrll 			lwp_unlock_to(l2, spc->spc_lwplock);
    137  1.50     skrll 		} else {
    138  1.50     skrll 			KASSERT(lwp_locked(l2, spc->spc_mutex));
    139   1.2        ad 			l2->l_stat = LSRUN;
    140  1.19      yamt 			sched_enqueue(l2, false);
    141  1.50     skrll 			lwp_unlock(l2);
    142   1.2        ad 		}
    143  1.31        ad 	} else {
    144   1.2        ad 		l2->l_stat = LSSUSPENDED;
    145  1.51      yamt 		p->p_nrlwps--;
    146  1.50     skrll 		lwp_unlock_to(l2, spc->spc_lwplock);
    147  1.31        ad 	}
    148  1.39        ad 	mutex_exit(p->p_lock);
    149   1.2        ad 
    150   1.2        ad 	return 0;
    151   1.2        ad }
    152   1.2        ad 
    153   1.2        ad int
    154  1.54    martin sys__lwp_create(struct lwp *l, const struct sys__lwp_create_args *uap,
    155  1.54    martin     register_t *retval)
    156  1.54    martin {
    157  1.54    martin 	/* {
    158  1.54    martin 		syscallarg(const ucontext_t *) ucp;
    159  1.54    martin 		syscallarg(u_long) flags;
    160  1.54    martin 		syscallarg(lwpid_t *) new_lwp;
    161  1.54    martin 	} */
    162  1.54    martin 	struct proc *p = l->l_proc;
    163  1.57      maxv 	ucontext_t *newuc;
    164  1.54    martin 	lwpid_t lid;
    165  1.54    martin 	int error;
    166  1.54    martin 
    167  1.54    martin 	newuc = kmem_alloc(sizeof(ucontext_t), KM_SLEEP);
    168  1.54    martin 	error = copyin(SCARG(uap, ucp), newuc, p->p_emul->e_ucsize);
    169  1.54    martin 	if (error)
    170  1.54    martin 		goto fail;
    171  1.54    martin 
    172  1.54    martin 	/* validate the ucontext */
    173  1.54    martin 	if ((newuc->uc_flags & _UC_CPU) == 0) {
    174  1.54    martin 		error = EINVAL;
    175  1.54    martin 		goto fail;
    176  1.54    martin 	}
    177  1.54    martin 	error = cpu_mcontext_validate(l, &newuc->uc_mcontext);
    178  1.54    martin 	if (error)
    179  1.54    martin 		goto fail;
    180  1.54    martin 
    181  1.59  christos 	const sigset_t *sigmask = newuc->uc_flags & _UC_SIGMASK ?
    182  1.59  christos 	    &newuc->uc_sigmask : &l->l_sigmask;
    183  1.59  christos 	error = do_lwp_create(l, newuc, SCARG(uap, flags), &lid, sigmask,
    184  1.59  christos 	    &SS_INIT);
    185  1.54    martin 	if (error)
    186  1.54    martin 		goto fail;
    187  1.54    martin 
    188  1.54    martin 	/*
    189  1.54    martin 	 * do not free ucontext in case of an error here,
    190  1.54    martin 	 * the lwp will actually run and access it
    191  1.54    martin 	 */
    192  1.54    martin 	return copyout(&lid, SCARG(uap, new_lwp), sizeof(lid));
    193  1.54    martin 
    194  1.54    martin fail:
    195  1.54    martin 	kmem_free(newuc, sizeof(ucontext_t));
    196  1.54    martin 	return error;
    197  1.54    martin }
    198  1.54    martin 
    199  1.54    martin int
    200  1.32       dsl sys__lwp_exit(struct lwp *l, const void *v, register_t *retval)
    201   1.2        ad {
    202   1.2        ad 
    203   1.2        ad 	lwp_exit(l);
    204   1.2        ad 	return 0;
    205   1.2        ad }
    206   1.2        ad 
    207   1.2        ad int
    208  1.32       dsl sys__lwp_self(struct lwp *l, const void *v, register_t *retval)
    209   1.2        ad {
    210   1.2        ad 
    211   1.2        ad 	*retval = l->l_lid;
    212   1.2        ad 	return 0;
    213   1.2        ad }
    214   1.2        ad 
    215   1.2        ad int
    216  1.32       dsl sys__lwp_getprivate(struct lwp *l, const void *v, register_t *retval)
    217   1.2        ad {
    218   1.2        ad 
    219   1.2        ad 	*retval = (uintptr_t)l->l_private;
    220   1.2        ad 	return 0;
    221   1.2        ad }
    222   1.2        ad 
    223   1.2        ad int
    224  1.47     rmind sys__lwp_setprivate(struct lwp *l, const struct sys__lwp_setprivate_args *uap,
    225  1.47     rmind     register_t *retval)
    226   1.2        ad {
    227  1.32       dsl 	/* {
    228   1.2        ad 		syscallarg(void *) ptr;
    229  1.32       dsl 	} */
    230   1.2        ad 
    231  1.52       chs 	return lwp_setprivate(l, SCARG(uap, ptr));
    232   1.2        ad }
    233   1.2        ad 
    234   1.2        ad int
    235  1.47     rmind sys__lwp_suspend(struct lwp *l, const struct sys__lwp_suspend_args *uap,
    236  1.47     rmind     register_t *retval)
    237   1.2        ad {
    238  1.32       dsl 	/* {
    239   1.2        ad 		syscallarg(lwpid_t) target;
    240  1.32       dsl 	} */
    241   1.2        ad 	struct proc *p = l->l_proc;
    242   1.2        ad 	struct lwp *t;
    243   1.2        ad 	int error;
    244   1.2        ad 
    245  1.39        ad 	mutex_enter(p->p_lock);
    246   1.2        ad 	if ((t = lwp_find(p, SCARG(uap, target))) == NULL) {
    247  1.39        ad 		mutex_exit(p->p_lock);
    248   1.2        ad 		return ESRCH;
    249   1.2        ad 	}
    250   1.2        ad 
    251   1.2        ad 	/*
    252   1.2        ad 	 * Check for deadlock, which is only possible when we're suspending
    253   1.2        ad 	 * ourself.  XXX There is a short race here, as p_nrlwps is only
    254   1.2        ad 	 * incremented when an LWP suspends itself on the kernel/user
    255   1.2        ad 	 * boundary.  It's still possible to kill -9 the process so we
    256   1.2        ad 	 * don't bother checking further.
    257   1.2        ad 	 */
    258   1.2        ad 	lwp_lock(t);
    259   1.2        ad 	if ((t == l && p->p_nrlwps == 1) ||
    260   1.4     pavel 	    (l->l_flag & (LW_WCORE | LW_WEXIT)) != 0) {
    261   1.2        ad 		lwp_unlock(t);
    262  1.39        ad 		mutex_exit(p->p_lock);
    263   1.2        ad 		return EDEADLK;
    264   1.2        ad 	}
    265   1.2        ad 
    266   1.2        ad 	/*
    267   1.2        ad 	 * Suspend the LWP.  XXX If it's on a different CPU, we should wait
    268   1.2        ad 	 * for it to be preempted, where it will put itself to sleep.
    269   1.2        ad 	 *
    270   1.2        ad 	 * Suspension of the current LWP will happen on return to userspace.
    271   1.2        ad 	 */
    272   1.2        ad 	error = lwp_suspend(l, t);
    273  1.23     rmind 	if (error) {
    274  1.39        ad 		mutex_exit(p->p_lock);
    275  1.23     rmind 		return error;
    276  1.23     rmind 	}
    277  1.23     rmind 
    278  1.23     rmind 	/*
    279  1.23     rmind 	 * Wait for:
    280  1.23     rmind 	 *  o process exiting
    281  1.23     rmind 	 *  o target LWP suspended
    282  1.23     rmind 	 *  o target LWP not suspended and L_WSUSPEND clear
    283  1.23     rmind 	 *  o target LWP exited
    284  1.23     rmind 	 */
    285  1.23     rmind 	for (;;) {
    286  1.39        ad 		error = cv_wait_sig(&p->p_lwpcv, p->p_lock);
    287  1.23     rmind 		if (error) {
    288  1.23     rmind 			error = ERESTART;
    289  1.23     rmind 			break;
    290  1.23     rmind 		}
    291  1.25     rmind 		if (lwp_find(p, SCARG(uap, target)) == NULL) {
    292  1.25     rmind 			error = ESRCH;
    293  1.25     rmind 			break;
    294  1.25     rmind 		}
    295  1.23     rmind 		if ((l->l_flag | t->l_flag) & (LW_WCORE | LW_WEXIT)) {
    296  1.23     rmind 			error = ERESTART;
    297  1.23     rmind 			break;
    298  1.23     rmind 		}
    299  1.23     rmind 		if (t->l_stat == LSSUSPENDED ||
    300  1.23     rmind 		    (t->l_flag & LW_WSUSPEND) == 0)
    301  1.23     rmind 			break;
    302  1.23     rmind 	}
    303  1.39        ad 	mutex_exit(p->p_lock);
    304   1.2        ad 
    305   1.2        ad 	return error;
    306   1.2        ad }
    307   1.2        ad 
    308   1.2        ad int
    309  1.47     rmind sys__lwp_continue(struct lwp *l, const struct sys__lwp_continue_args *uap,
    310  1.47     rmind     register_t *retval)
    311   1.2        ad {
    312  1.32       dsl 	/* {
    313   1.2        ad 		syscallarg(lwpid_t) target;
    314  1.32       dsl 	} */
    315   1.2        ad 	int error;
    316   1.2        ad 	struct proc *p = l->l_proc;
    317   1.2        ad 	struct lwp *t;
    318   1.2        ad 
    319   1.2        ad 	error = 0;
    320   1.2        ad 
    321  1.39        ad 	mutex_enter(p->p_lock);
    322   1.2        ad 	if ((t = lwp_find(p, SCARG(uap, target))) == NULL) {
    323  1.39        ad 		mutex_exit(p->p_lock);
    324   1.2        ad 		return ESRCH;
    325   1.2        ad 	}
    326   1.2        ad 
    327   1.2        ad 	lwp_lock(t);
    328   1.2        ad 	lwp_continue(t);
    329  1.39        ad 	mutex_exit(p->p_lock);
    330   1.2        ad 
    331   1.2        ad 	return error;
    332   1.2        ad }
    333   1.2        ad 
    334   1.2        ad int
    335  1.47     rmind sys__lwp_wakeup(struct lwp *l, const struct sys__lwp_wakeup_args *uap,
    336  1.47     rmind     register_t *retval)
    337   1.2        ad {
    338  1.32       dsl 	/* {
    339   1.2        ad 		syscallarg(lwpid_t) target;
    340  1.32       dsl 	} */
    341   1.2        ad 	struct lwp *t;
    342   1.2        ad 	struct proc *p;
    343   1.2        ad 	int error;
    344   1.2        ad 
    345   1.2        ad 	p = l->l_proc;
    346  1.39        ad 	mutex_enter(p->p_lock);
    347   1.2        ad 
    348   1.2        ad 	if ((t = lwp_find(p, SCARG(uap, target))) == NULL) {
    349  1.39        ad 		mutex_exit(p->p_lock);
    350   1.2        ad 		return ESRCH;
    351   1.2        ad 	}
    352   1.2        ad 
    353   1.2        ad 	lwp_lock(t);
    354  1.15        ad 	t->l_flag |= (LW_CANCELLED | LW_UNPARKED);
    355   1.2        ad 
    356   1.2        ad 	if (t->l_stat != LSSLEEP) {
    357  1.16        ad 		lwp_unlock(t);
    358   1.2        ad 		error = ENODEV;
    359  1.16        ad 	} else if ((t->l_flag & LW_SINTR) == 0) {
    360  1.16        ad 		lwp_unlock(t);
    361   1.2        ad 		error = EBUSY;
    362  1.16        ad 	} else {
    363  1.16        ad 		/* Wake it up.  lwp_unsleep() will release the LWP lock. */
    364  1.46     rmind 		lwp_unsleep(t, true);
    365  1.16        ad 		error = 0;
    366   1.2        ad 	}
    367   1.2        ad 
    368  1.39        ad 	mutex_exit(p->p_lock);
    369   1.2        ad 
    370   1.2        ad 	return error;
    371   1.2        ad }
    372   1.2        ad 
    373   1.2        ad int
    374  1.47     rmind sys__lwp_wait(struct lwp *l, const struct sys__lwp_wait_args *uap,
    375  1.47     rmind     register_t *retval)
    376   1.2        ad {
    377  1.32       dsl 	/* {
    378   1.2        ad 		syscallarg(lwpid_t) wait_for;
    379   1.2        ad 		syscallarg(lwpid_t *) departed;
    380  1.32       dsl 	} */
    381   1.2        ad 	struct proc *p = l->l_proc;
    382   1.2        ad 	int error;
    383   1.2        ad 	lwpid_t dep;
    384   1.2        ad 
    385  1.39        ad 	mutex_enter(p->p_lock);
    386  1.55     rmind 	error = lwp_wait(l, SCARG(uap, wait_for), &dep, false);
    387  1.39        ad 	mutex_exit(p->p_lock);
    388   1.2        ad 
    389  1.55     rmind 	if (!error && SCARG(uap, departed)) {
    390   1.2        ad 		error = copyout(&dep, SCARG(uap, departed), sizeof(dep));
    391   1.2        ad 	}
    392   1.2        ad 
    393  1.55     rmind 	return error;
    394   1.2        ad }
    395   1.2        ad 
    396   1.2        ad int
    397  1.47     rmind sys__lwp_kill(struct lwp *l, const struct sys__lwp_kill_args *uap,
    398  1.47     rmind     register_t *retval)
    399   1.2        ad {
    400  1.32       dsl 	/* {
    401   1.2        ad 		syscallarg(lwpid_t)	target;
    402   1.2        ad 		syscallarg(int)		signo;
    403  1.32       dsl 	} */
    404   1.2        ad 	struct proc *p = l->l_proc;
    405   1.2        ad 	struct lwp *t;
    406   1.2        ad 	ksiginfo_t ksi;
    407   1.2        ad 	int signo = SCARG(uap, signo);
    408   1.2        ad 	int error = 0;
    409   1.2        ad 
    410   1.2        ad 	if ((u_int)signo >= NSIG)
    411   1.2        ad 		return EINVAL;
    412   1.2        ad 
    413   1.2        ad 	KSI_INIT(&ksi);
    414   1.2        ad 	ksi.ksi_signo = signo;
    415  1.43        ad 	ksi.ksi_code = SI_LWP;
    416   1.2        ad 	ksi.ksi_pid = p->p_pid;
    417   1.2        ad 	ksi.ksi_uid = kauth_cred_geteuid(l->l_cred);
    418   1.2        ad 	ksi.ksi_lid = SCARG(uap, target);
    419   1.2        ad 
    420  1.38        ad 	mutex_enter(proc_lock);
    421  1.39        ad 	mutex_enter(p->p_lock);
    422   1.2        ad 	if ((t = lwp_find(p, ksi.ksi_lid)) == NULL)
    423   1.2        ad 		error = ESRCH;
    424   1.2        ad 	else if (signo != 0)
    425   1.2        ad 		kpsignal2(p, &ksi);
    426  1.39        ad 	mutex_exit(p->p_lock);
    427  1.38        ad 	mutex_exit(proc_lock);
    428   1.2        ad 
    429   1.2        ad 	return error;
    430   1.2        ad }
    431   1.2        ad 
    432   1.2        ad int
    433  1.47     rmind sys__lwp_detach(struct lwp *l, const struct sys__lwp_detach_args *uap,
    434  1.47     rmind     register_t *retval)
    435   1.2        ad {
    436  1.32       dsl 	/* {
    437   1.2        ad 		syscallarg(lwpid_t)	target;
    438  1.32       dsl 	} */
    439   1.2        ad 	struct proc *p;
    440   1.2        ad 	struct lwp *t;
    441   1.2        ad 	lwpid_t target;
    442   1.2        ad 	int error;
    443   1.2        ad 
    444   1.2        ad 	target = SCARG(uap, target);
    445   1.2        ad 	p = l->l_proc;
    446   1.2        ad 
    447  1.39        ad 	mutex_enter(p->p_lock);
    448   1.2        ad 
    449   1.2        ad 	if (l->l_lid == target)
    450   1.2        ad 		t = l;
    451   1.2        ad 	else {
    452   1.2        ad 		/*
    453   1.2        ad 		 * We can't use lwp_find() here because the target might
    454   1.2        ad 		 * be a zombie.
    455   1.2        ad 		 */
    456   1.2        ad 		LIST_FOREACH(t, &p->p_lwps, l_sibling)
    457   1.2        ad 			if (t->l_lid == target)
    458   1.2        ad 				break;
    459   1.2        ad 	}
    460   1.2        ad 
    461   1.2        ad 	/*
    462   1.2        ad 	 * If the LWP is already detached, there's nothing to do.
    463   1.2        ad 	 * If it's a zombie, we need to clean up after it.  LSZOMB
    464   1.2        ad 	 * is visible with the proc mutex held.
    465   1.2        ad 	 *
    466   1.2        ad 	 * After we have detached or released the LWP, kick any
    467   1.2        ad 	 * other LWPs that may be sitting in _lwp_wait(), waiting
    468   1.2        ad 	 * for the target LWP to exit.
    469   1.2        ad 	 */
    470   1.2        ad 	if (t != NULL && t->l_stat != LSIDL) {
    471   1.2        ad 		if ((t->l_prflag & LPR_DETACHED) == 0) {
    472   1.2        ad 			p->p_ndlwps++;
    473   1.2        ad 			t->l_prflag |= LPR_DETACHED;
    474   1.2        ad 			if (t->l_stat == LSZOMB) {
    475  1.17        ad 				/* Releases proc mutex. */
    476  1.17        ad 				lwp_free(t, false, false);
    477   1.2        ad 				return 0;
    478   1.2        ad 			}
    479   1.2        ad 			error = 0;
    480  1.17        ad 
    481  1.17        ad 			/*
    482  1.17        ad 			 * Have any LWPs sleeping in lwp_wait() recheck
    483  1.17        ad 			 * for deadlock.
    484  1.17        ad 			 */
    485  1.17        ad 			cv_broadcast(&p->p_lwpcv);
    486   1.2        ad 		} else
    487   1.2        ad 			error = EINVAL;
    488   1.2        ad 	} else
    489   1.2        ad 		error = ESRCH;
    490   1.2        ad 
    491  1.39        ad 	mutex_exit(p->p_lock);
    492   1.2        ad 
    493   1.2        ad 	return error;
    494   1.2        ad }
    495   1.2        ad 
    496   1.2        ad static inline wchan_t
    497   1.2        ad lwp_park_wchan(struct proc *p, const void *hint)
    498   1.2        ad {
    499  1.22        ad 
    500   1.2        ad 	return (wchan_t)((uintptr_t)p ^ (uintptr_t)hint);
    501   1.2        ad }
    502   1.2        ad 
    503   1.2        ad int
    504  1.24        ad lwp_unpark(lwpid_t target, const void *hint)
    505   1.2        ad {
    506  1.24        ad 	sleepq_t *sq;
    507  1.24        ad 	wchan_t wchan;
    508  1.41        ad 	kmutex_t *mp;
    509  1.24        ad 	proc_t *p;
    510  1.24        ad 	lwp_t *t;
    511  1.24        ad 
    512  1.24        ad 	/*
    513  1.24        ad 	 * Easy case: search for the LWP on the sleep queue.  If
    514  1.24        ad 	 * it's parked, remove it from the queue and set running.
    515  1.24        ad 	 */
    516  1.24        ad 	p = curproc;
    517  1.24        ad 	wchan = lwp_park_wchan(p, hint);
    518  1.41        ad 	sq = sleeptab_lookup(&lwp_park_tab, wchan, &mp);
    519  1.24        ad 
    520  1.41        ad 	TAILQ_FOREACH(t, sq, l_sleepchain)
    521  1.24        ad 		if (t->l_proc == p && t->l_lid == target)
    522  1.24        ad 			break;
    523  1.24        ad 
    524  1.24        ad 	if (__predict_true(t != NULL)) {
    525  1.46     rmind 		sleepq_remove(sq, t);
    526  1.41        ad 		mutex_spin_exit(mp);
    527  1.24        ad 		return 0;
    528  1.24        ad 	}
    529  1.24        ad 
    530  1.24        ad 	/*
    531  1.24        ad 	 * The LWP hasn't parked yet.  Take the hit and mark the
    532  1.24        ad 	 * operation as pending.
    533  1.24        ad 	 */
    534  1.41        ad 	mutex_spin_exit(mp);
    535  1.20       dsl 
    536  1.39        ad 	mutex_enter(p->p_lock);
    537  1.24        ad 	if ((t = lwp_find(p, target)) == NULL) {
    538  1.39        ad 		mutex_exit(p->p_lock);
    539  1.24        ad 		return ESRCH;
    540  1.24        ad 	}
    541  1.20       dsl 
    542  1.24        ad 	/*
    543  1.24        ad 	 * It may not have parked yet, we may have raced, or it
    544  1.24        ad 	 * is parked on a different user sync object.
    545  1.24        ad 	 */
    546  1.24        ad 	lwp_lock(t);
    547  1.24        ad 	if (t->l_syncobj == &lwp_park_sobj) {
    548  1.24        ad 		/* Releases the LWP lock. */
    549  1.46     rmind 		lwp_unsleep(t, true);
    550  1.24        ad 	} else {
    551  1.24        ad 		/*
    552  1.24        ad 		 * Set the operation pending.  The next call to _lwp_park
    553  1.24        ad 		 * will return early.
    554  1.24        ad 		 */
    555  1.24        ad 		t->l_flag |= LW_UNPARKED;
    556  1.24        ad 		lwp_unlock(t);
    557  1.24        ad 	}
    558  1.20       dsl 
    559  1.39        ad 	mutex_exit(p->p_lock);
    560  1.24        ad 	return 0;
    561  1.20       dsl }
    562  1.20       dsl 
    563  1.20       dsl int
    564  1.56  christos lwp_park(clockid_t clock_id, int flags, struct timespec *ts, const void *hint)
    565  1.20       dsl {
    566   1.2        ad 	sleepq_t *sq;
    567  1.41        ad 	kmutex_t *mp;
    568   1.2        ad 	wchan_t wchan;
    569   1.2        ad 	int timo, error;
    570  1.62  christos 	struct timespec start;
    571  1.24        ad 	lwp_t *l;
    572  1.62  christos 	bool timeremain = !(flags & TIMER_ABSTIME) && ts;
    573   1.2        ad 
    574  1.20       dsl 	if (ts != NULL) {
    575  1.62  christos 		if ((error = ts2timo(clock_id, flags, ts, &timo,
    576  1.62  christos 		    timeremain ? &start : NULL)) != 0)
    577   1.2        ad 			return error;
    578  1.24        ad 		KASSERT(timo != 0);
    579  1.48     rmind 	} else {
    580   1.2        ad 		timo = 0;
    581  1.48     rmind 	}
    582   1.2        ad 
    583   1.2        ad 	/* Find and lock the sleep queue. */
    584  1.24        ad 	l = curlwp;
    585  1.20       dsl 	wchan = lwp_park_wchan(l->l_proc, hint);
    586  1.41        ad 	sq = sleeptab_lookup(&lwp_park_tab, wchan, &mp);
    587   1.2        ad 
    588   1.2        ad 	/*
    589   1.2        ad 	 * Before going the full route and blocking, check to see if an
    590   1.2        ad 	 * unpark op is pending.
    591   1.2        ad 	 */
    592  1.19      yamt 	lwp_lock(l);
    593   1.8        ad 	if ((l->l_flag & (LW_CANCELLED | LW_UNPARKED)) != 0) {
    594   1.8        ad 		l->l_flag &= ~(LW_CANCELLED | LW_UNPARKED);
    595  1.19      yamt 		lwp_unlock(l);
    596  1.41        ad 		mutex_spin_exit(mp);
    597   1.2        ad 		return EALREADY;
    598   1.2        ad 	}
    599  1.41        ad 	lwp_unlock_to(l, mp);
    600  1.24        ad 	l->l_biglocks = 0;
    601  1.27        ad 	sleepq_enqueue(sq, wchan, "parked", &lwp_park_sobj);
    602  1.19      yamt 	error = sleepq_block(timo, true);
    603  1.13      yamt 	switch (error) {
    604  1.14      yamt 	case EWOULDBLOCK:
    605  1.14      yamt 		error = ETIMEDOUT;
    606  1.62  christos 		if (timeremain)
    607  1.62  christos 			memset(ts, 0, sizeof(*ts));
    608  1.14      yamt 		break;
    609  1.14      yamt 	case ERESTART:
    610  1.14      yamt 		error = EINTR;
    611  1.62  christos 		/*FALLTHROUGH*/
    612  1.14      yamt 	default:
    613  1.62  christos 		if (timeremain)
    614  1.62  christos 			clock_timeleft(clock_id, ts, &start);
    615  1.14      yamt 		break;
    616  1.13      yamt 	}
    617  1.13      yamt 	return error;
    618   1.2        ad }
    619   1.2        ad 
    620  1.24        ad /*
    621  1.24        ad  * 'park' an LWP waiting on a user-level synchronisation object.  The LWP
    622  1.24        ad  * will remain parked until another LWP in the same process calls in and
    623  1.24        ad  * requests that it be unparked.
    624  1.24        ad  */
    625   1.2        ad int
    626  1.56  christos sys____lwp_park60(struct lwp *l, const struct sys____lwp_park60_args *uap,
    627  1.44  christos     register_t *retval)
    628   1.2        ad {
    629  1.32       dsl 	/* {
    630  1.56  christos 		syscallarg(clockid_t)			clock_id;
    631  1.56  christos 		syscallarg(int)				flags;
    632  1.62  christos 		syscallarg(struct timespec *)		ts;
    633  1.24        ad 		syscallarg(lwpid_t)			unpark;
    634  1.24        ad 		syscallarg(const void *)		hint;
    635  1.24        ad 		syscallarg(const void *)		unparkhint;
    636  1.32       dsl 	} */
    637  1.24        ad 	struct timespec ts, *tsp;
    638  1.24        ad 	int error;
    639   1.2        ad 
    640  1.24        ad 	if (SCARG(uap, ts) == NULL)
    641  1.24        ad 		tsp = NULL;
    642  1.24        ad 	else {
    643  1.24        ad 		error = copyin(SCARG(uap, ts), &ts, sizeof(ts));
    644  1.24        ad 		if (error != 0)
    645  1.24        ad 			return error;
    646  1.24        ad 		tsp = &ts;
    647  1.24        ad 	}
    648   1.2        ad 
    649  1.24        ad 	if (SCARG(uap, unpark) != 0) {
    650  1.24        ad 		error = lwp_unpark(SCARG(uap, unpark), SCARG(uap, unparkhint));
    651  1.24        ad 		if (error != 0)
    652  1.24        ad 			return error;
    653  1.15        ad 	}
    654  1.15        ad 
    655  1.62  christos 	error = lwp_park(SCARG(uap, clock_id), SCARG(uap, flags), tsp,
    656  1.56  christos 	    SCARG(uap, hint));
    657  1.62  christos 	if (SCARG(uap, ts) != NULL && (SCARG(uap, flags) & TIMER_ABSTIME) == 0)
    658  1.62  christos 		(void)copyout(tsp, SCARG(uap, ts), sizeof(*tsp));
    659  1.62  christos 	return error;
    660  1.24        ad }
    661   1.2        ad 
    662  1.24        ad int
    663  1.47     rmind sys__lwp_unpark(struct lwp *l, const struct sys__lwp_unpark_args *uap,
    664  1.47     rmind     register_t *retval)
    665  1.24        ad {
    666  1.32       dsl 	/* {
    667  1.24        ad 		syscallarg(lwpid_t)		target;
    668  1.24        ad 		syscallarg(const void *)	hint;
    669  1.32       dsl 	} */
    670   1.2        ad 
    671  1.24        ad 	return lwp_unpark(SCARG(uap, target), SCARG(uap, hint));
    672   1.2        ad }
    673   1.2        ad 
    674   1.2        ad int
    675  1.47     rmind sys__lwp_unpark_all(struct lwp *l, const struct sys__lwp_unpark_all_args *uap,
    676  1.47     rmind     register_t *retval)
    677   1.2        ad {
    678  1.32       dsl 	/* {
    679   1.2        ad 		syscallarg(const lwpid_t *)	targets;
    680   1.2        ad 		syscallarg(size_t)		ntargets;
    681   1.2        ad 		syscallarg(const void *)	hint;
    682  1.32       dsl 	} */
    683   1.2        ad 	struct proc *p;
    684   1.2        ad 	struct lwp *t;
    685   1.2        ad 	sleepq_t *sq;
    686   1.2        ad 	wchan_t wchan;
    687   1.2        ad 	lwpid_t targets[32], *tp, *tpp, *tmax, target;
    688  1.46     rmind 	int error;
    689  1.41        ad 	kmutex_t *mp;
    690  1.15        ad 	u_int ntargets;
    691   1.2        ad 	size_t sz;
    692   1.2        ad 
    693   1.2        ad 	p = l->l_proc;
    694   1.2        ad 	ntargets = SCARG(uap, ntargets);
    695   1.2        ad 
    696   1.2        ad 	if (SCARG(uap, targets) == NULL) {
    697   1.2        ad 		/*
    698   1.2        ad 		 * Let the caller know how much we are willing to do, and
    699   1.2        ad 		 * let it unpark the LWPs in blocks.
    700   1.2        ad 		 */
    701   1.2        ad 		*retval = LWP_UNPARK_MAX;
    702   1.2        ad 		return 0;
    703   1.2        ad 	}
    704   1.2        ad 	if (ntargets > LWP_UNPARK_MAX || ntargets == 0)
    705   1.2        ad 		return EINVAL;
    706   1.2        ad 
    707   1.2        ad 	/*
    708   1.2        ad 	 * Copy in the target array.  If it's a small number of LWPs, then
    709   1.2        ad 	 * place the numbers on the stack.
    710   1.2        ad 	 */
    711   1.2        ad 	sz = sizeof(target) * ntargets;
    712   1.2        ad 	if (sz <= sizeof(targets))
    713   1.2        ad 		tp = targets;
    714  1.61       chs 	else
    715   1.2        ad 		tp = kmem_alloc(sz, KM_SLEEP);
    716   1.2        ad 	error = copyin(SCARG(uap, targets), tp, sz);
    717   1.2        ad 	if (error != 0) {
    718   1.2        ad 		if (tp != targets) {
    719   1.2        ad 			kmem_free(tp, sz);
    720   1.2        ad 		}
    721   1.2        ad 		return error;
    722   1.2        ad 	}
    723   1.2        ad 
    724   1.2        ad 	wchan = lwp_park_wchan(p, SCARG(uap, hint));
    725  1.41        ad 	sq = sleeptab_lookup(&lwp_park_tab, wchan, &mp);
    726   1.2        ad 
    727   1.2        ad 	for (tmax = tp + ntargets, tpp = tp; tpp < tmax; tpp++) {
    728   1.2        ad 		target = *tpp;
    729   1.2        ad 
    730   1.2        ad 		/*
    731   1.2        ad 		 * Easy case: search for the LWP on the sleep queue.  If
    732   1.2        ad 		 * it's parked, remove it from the queue and set running.
    733   1.2        ad 		 */
    734  1.41        ad 		TAILQ_FOREACH(t, sq, l_sleepchain)
    735   1.2        ad 			if (t->l_proc == p && t->l_lid == target)
    736   1.2        ad 				break;
    737   1.2        ad 
    738   1.2        ad 		if (t != NULL) {
    739  1.46     rmind 			sleepq_remove(sq, t);
    740   1.2        ad 			continue;
    741   1.2        ad 		}
    742   1.2        ad 
    743   1.2        ad 		/*
    744   1.2        ad 		 * The LWP hasn't parked yet.  Take the hit and
    745   1.2        ad 		 * mark the operation as pending.
    746   1.2        ad 		 */
    747  1.41        ad 		mutex_spin_exit(mp);
    748  1.39        ad 		mutex_enter(p->p_lock);
    749   1.2        ad 		if ((t = lwp_find(p, target)) == NULL) {
    750  1.39        ad 			mutex_exit(p->p_lock);
    751  1.41        ad 			mutex_spin_enter(mp);
    752   1.2        ad 			continue;
    753   1.2        ad 		}
    754   1.2        ad 		lwp_lock(t);
    755   1.2        ad 
    756  1.15        ad 		/*
    757  1.15        ad 		 * It may not have parked yet, we may have raced, or
    758  1.15        ad 		 * it is parked on a different user sync object.
    759  1.15        ad 		 */
    760  1.15        ad 		if (t->l_syncobj == &lwp_park_sobj) {
    761  1.15        ad 			/* Releases the LWP lock. */
    762  1.46     rmind 			lwp_unsleep(t, true);
    763   1.2        ad 		} else {
    764   1.2        ad 			/*
    765  1.15        ad 			 * Set the operation pending.  The next call to
    766  1.15        ad 			 * _lwp_park will return early.
    767   1.2        ad 			 */
    768   1.8        ad 			t->l_flag |= LW_UNPARKED;
    769   1.2        ad 			lwp_unlock(t);
    770   1.2        ad 		}
    771  1.15        ad 
    772  1.39        ad 		mutex_exit(p->p_lock);
    773  1.41        ad 		mutex_spin_enter(mp);
    774   1.2        ad 	}
    775   1.2        ad 
    776  1.41        ad 	mutex_spin_exit(mp);
    777  1.33        ad 	if (tp != targets)
    778   1.2        ad 		kmem_free(tp, sz);
    779  1.15        ad 
    780   1.2        ad 	return 0;
    781   1.2        ad }
    782  1.28        ad 
    783  1.28        ad int
    784  1.47     rmind sys__lwp_setname(struct lwp *l, const struct sys__lwp_setname_args *uap,
    785  1.47     rmind     register_t *retval)
    786  1.28        ad {
    787  1.32       dsl 	/* {
    788  1.28        ad 		syscallarg(lwpid_t)		target;
    789  1.28        ad 		syscallarg(const char *)	name;
    790  1.32       dsl 	} */
    791  1.28        ad 	char *name, *oname;
    792  1.30        ad 	lwpid_t target;
    793  1.28        ad 	proc_t *p;
    794  1.28        ad 	lwp_t *t;
    795  1.28        ad 	int error;
    796  1.28        ad 
    797  1.30        ad 	if ((target = SCARG(uap, target)) == 0)
    798  1.30        ad 		target = l->l_lid;
    799  1.30        ad 
    800  1.28        ad 	name = kmem_alloc(MAXCOMLEN, KM_SLEEP);
    801  1.28        ad 	error = copyinstr(SCARG(uap, name), name, MAXCOMLEN, NULL);
    802  1.28        ad 	switch (error) {
    803  1.28        ad 	case ENAMETOOLONG:
    804  1.28        ad 	case 0:
    805  1.28        ad 		name[MAXCOMLEN - 1] = '\0';
    806  1.28        ad 		break;
    807  1.28        ad 	default:
    808  1.28        ad 		kmem_free(name, MAXCOMLEN);
    809  1.28        ad 		return error;
    810  1.28        ad 	}
    811  1.28        ad 
    812  1.28        ad 	p = curproc;
    813  1.39        ad 	mutex_enter(p->p_lock);
    814  1.30        ad 	if ((t = lwp_find(p, target)) == NULL) {
    815  1.39        ad 		mutex_exit(p->p_lock);
    816  1.28        ad 		kmem_free(name, MAXCOMLEN);
    817  1.28        ad 		return ESRCH;
    818  1.28        ad 	}
    819  1.28        ad 	lwp_lock(t);
    820  1.28        ad 	oname = t->l_name;
    821  1.28        ad 	t->l_name = name;
    822  1.28        ad 	lwp_unlock(t);
    823  1.39        ad 	mutex_exit(p->p_lock);
    824  1.28        ad 
    825  1.28        ad 	if (oname != NULL)
    826  1.28        ad 		kmem_free(oname, MAXCOMLEN);
    827  1.28        ad 
    828  1.28        ad 	return 0;
    829  1.28        ad }
    830  1.28        ad 
    831  1.28        ad int
    832  1.47     rmind sys__lwp_getname(struct lwp *l, const struct sys__lwp_getname_args *uap,
    833  1.47     rmind     register_t *retval)
    834  1.28        ad {
    835  1.32       dsl 	/* {
    836  1.28        ad 		syscallarg(lwpid_t)		target;
    837  1.28        ad 		syscallarg(char *)		name;
    838  1.28        ad 		syscallarg(size_t)		len;
    839  1.32       dsl 	} */
    840  1.28        ad 	char name[MAXCOMLEN];
    841  1.30        ad 	lwpid_t target;
    842  1.68      maxv 	size_t len;
    843  1.28        ad 	proc_t *p;
    844  1.28        ad 	lwp_t *t;
    845  1.28        ad 
    846  1.30        ad 	if ((target = SCARG(uap, target)) == 0)
    847  1.30        ad 		target = l->l_lid;
    848  1.30        ad 
    849  1.28        ad 	p = curproc;
    850  1.39        ad 	mutex_enter(p->p_lock);
    851  1.30        ad 	if ((t = lwp_find(p, target)) == NULL) {
    852  1.39        ad 		mutex_exit(p->p_lock);
    853  1.28        ad 		return ESRCH;
    854  1.28        ad 	}
    855  1.28        ad 	lwp_lock(t);
    856  1.28        ad 	if (t->l_name == NULL)
    857  1.28        ad 		name[0] = '\0';
    858  1.28        ad 	else
    859  1.58      maya 		strlcpy(name, t->l_name, sizeof(name));
    860  1.28        ad 	lwp_unlock(t);
    861  1.39        ad 	mutex_exit(p->p_lock);
    862  1.28        ad 
    863  1.68      maxv 	len = uimin(SCARG(uap, len), sizeof(name));
    864  1.68      maxv 
    865  1.68      maxv 	return copyoutstr(name, SCARG(uap, name), len, NULL);
    866  1.28        ad }
    867  1.30        ad 
    868  1.30        ad int
    869  1.47     rmind sys__lwp_ctl(struct lwp *l, const struct sys__lwp_ctl_args *uap,
    870  1.47     rmind     register_t *retval)
    871  1.30        ad {
    872  1.32       dsl 	/* {
    873  1.30        ad 		syscallarg(int)			features;
    874  1.30        ad 		syscallarg(struct lwpctl **)	address;
    875  1.32       dsl 	} */
    876  1.30        ad 	int error, features;
    877  1.30        ad 	vaddr_t vaddr;
    878  1.30        ad 
    879  1.30        ad 	features = SCARG(uap, features);
    880  1.35        ad 	features &= ~(LWPCTL_FEATURE_CURCPU | LWPCTL_FEATURE_PCTR);
    881  1.35        ad 	if (features != 0)
    882  1.30        ad 		return ENODEV;
    883  1.30        ad 	if ((error = lwp_ctl_alloc(&vaddr)) != 0)
    884  1.30        ad 		return error;
    885  1.30        ad 	return copyout(&vaddr, SCARG(uap, address), sizeof(void *));
    886  1.30        ad }
    887