1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 1999 Peter Wemm <peter@FreeBSD.org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include <sys/param.h>
30#include <sys/systm.h>
31#include <sys/cpuset.h>
32#include <sys/kthread.h>
33#include <sys/lock.h>
34#include <sys/mutex.h>
35#include <sys/proc.h>
36#include <sys/resourcevar.h>
37#include <sys/rwlock.h>
38#include <sys/signalvar.h>
39#include <sys/sysent.h>
40#include <sys/sx.h>
41#include <sys/umtxvar.h>
42#include <sys/unistd.h>
43#include <sys/wait.h>
44#include <sys/sched.h>
45#include <sys/tslog.h>
46#include <vm/vm.h>
47#include <vm/vm_extern.h>
48
49#include <machine/stdarg.h>
50
51/*
52 * Start a kernel process.  This is called after a fork() call in
53 * mi_startup() in the file kern/init_main.c.
54 *
55 * This function is used to start "internal" daemons and intended
56 * to be called from SYSINIT().
57 */
58void
59kproc_start(const void *udata)
60{
61	const struct kproc_desc	*kp = udata;
62	int error;
63
64	error = kproc_create((void (*)(void *))kp->func, NULL,
65	    kp->global_procpp, 0, 0, "%s", kp->arg0);
66	if (error)
67		panic("kproc_start: %s: error %d", kp->arg0, error);
68}
69
70/*
71 * Create a kernel process/thread/whatever.  It shares its address space
72 * with proc0 - ie: kernel only.
73 *
74 * func is the function to start.
75 * arg is the parameter to pass to function on first startup.
76 * newpp is the return value pointing to the thread's struct proc.
77 * flags are flags to fork1 (in unistd.h)
78 * fmt and following will be *printf'd into (*newpp)->p_comm (for ps, etc.).
79 */
80static int
81kproc_create1(void (*func)(void *), void *arg,
82    struct proc **newpp, int flags, int pages, const char *tdname)
83{
84	struct fork_req fr;
85	int error;
86	struct thread *td;
87	struct proc *p2;
88
89	if (!proc0.p_stats)
90		panic("kproc_create called too soon");
91
92	bzero(&fr, sizeof(fr));
93	fr.fr_flags = RFMEM | RFFDG | RFPROC | RFSTOPPED | flags;
94	fr.fr_flags2 = FR2_KPROC;
95	fr.fr_pages = pages;
96	fr.fr_procp = &p2;
97	error = fork1(&thread0, &fr);
98	if (error != 0)
99		return (error);
100
101	/* save a global descriptor, if desired */
102	if (newpp != NULL)
103		*newpp = p2;
104
105	/* set up arg0 for 'ps', et al */
106	strcpy(p2->p_comm, tdname);
107	td = FIRST_THREAD_IN_PROC(p2);
108	strcpy(td->td_name, tdname);
109#ifdef KTR
110	sched_clear_tdname(td);
111#endif
112	TSTHREAD(td, td->td_name);
113#ifdef HWPMC_HOOKS
114	if (PMC_SYSTEM_SAMPLING_ACTIVE()) {
115		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_PROC_CREATE_LOG, p2);
116		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_CREATE_LOG, NULL);
117	}
118#endif
119
120	/* call the processes' main()... */
121	cpu_fork_kthread_handler(td, func, arg);
122
123	/* Avoid inheriting affinity from a random parent. */
124	cpuset_kernthread(td);
125	thread_lock(td);
126	TD_SET_CAN_RUN(td);
127	sched_prio(td, PVM);
128	sched_user_prio(td, PUSER);
129
130	/* Delay putting it on the run queue until now. */
131	if ((flags & RFSTOPPED) == 0)
132		sched_add(td, SRQ_BORING);
133	else
134		thread_unlock(td);
135
136	return (0);
137}
138
139int
140kproc_create(void (*func)(void *), void *arg,
141    struct proc **newpp, int flags, int pages, const char *fmt, ...)
142{
143	va_list ap;
144	int error;
145	char tdname[MAXCOMLEN + 1];
146
147	va_start(ap, fmt);
148	vsnprintf(tdname, sizeof(tdname), fmt, ap);
149	va_end(ap);
150	DROP_GIANT();
151	error = kproc_create1(func, arg, newpp, flags, pages, tdname);
152	PICKUP_GIANT();
153	return (error);
154}
155
156void
157kproc_exit(int ecode)
158{
159	struct thread *td;
160	struct proc *p;
161
162	td = curthread;
163	p = td->td_proc;
164
165	/*
166	 * Reparent curthread from proc0 to init so that the zombie
167	 * is harvested.
168	 */
169	sx_xlock(&proctree_lock);
170	PROC_LOCK(p);
171	proc_reparent(p, initproc, true);
172	PROC_UNLOCK(p);
173	sx_xunlock(&proctree_lock);
174
175	/*
176	 * Wakeup anyone waiting for us to exit.
177	 */
178	wakeup(p);
179
180	/* Buh-bye! */
181	exit1(td, ecode, 0);
182}
183
184/*
185 * Advise a kernel process to suspend (or resume) in its main loop.
186 * Participation is voluntary.
187 */
188int
189kproc_suspend(struct proc *p, int timo)
190{
191	/*
192	 * Make sure this is indeed a system process and we can safely
193	 * use the p_siglist field.
194	 */
195	PROC_LOCK(p);
196	if ((p->p_flag & P_KPROC) == 0) {
197		PROC_UNLOCK(p);
198		return (EINVAL);
199	}
200	SIGADDSET(p->p_siglist, SIGSTOP);
201	wakeup(p);
202	return (msleep(&p->p_siglist, &p->p_mtx, PPAUSE | PDROP,
203	    "suspkp", timo));
204}
205
206int
207kproc_resume(struct proc *p)
208{
209	/*
210	 * Make sure this is indeed a system process and we can safely
211	 * use the p_siglist field.
212	 */
213	PROC_LOCK(p);
214	if ((p->p_flag & P_KPROC) == 0) {
215		PROC_UNLOCK(p);
216		return (EINVAL);
217	}
218	SIGDELSET(p->p_siglist, SIGSTOP);
219	PROC_UNLOCK(p);
220	wakeup(&p->p_siglist);
221	return (0);
222}
223
224void
225kproc_suspend_check(struct proc *p)
226{
227	PROC_LOCK(p);
228	while (SIGISMEMBER(p->p_siglist, SIGSTOP)) {
229		wakeup(&p->p_siglist);
230		msleep(&p->p_siglist, &p->p_mtx, PPAUSE, "kpsusp", 0);
231	}
232	PROC_UNLOCK(p);
233}
234
235/*
236 * Start a kernel thread.
237 *
238 * This function is used to start "internal" daemons and intended
239 * to be called from SYSINIT().
240 */
241
242void
243kthread_start(const void *udata)
244{
245	const struct kthread_desc *kp = udata;
246	int error;
247
248	error = kthread_add((void (*)(void *))kp->func, NULL,
249	    NULL, kp->global_threadpp, 0, 0, "%s", kp->arg0);
250	if (error)
251		panic("kthread_start: %s: error %d", kp->arg0, error);
252}
253
254/*
255 * Create a kernel thread.  It shares its address space
256 * with proc0 - ie: kernel only.
257 *
258 * func is the function to start.
259 * arg is the parameter to pass to function on first startup.
260 * newtdp is the return value pointing to the thread's struct thread.
261 *  ** XXX fix this --> flags are flags to fork1 (in unistd.h)
262 * fmt and following will be *printf'd into (*newtd)->td_name (for ps, etc.).
263 */
264static int
265kthread_add1(void (*func)(void *), void *arg, struct proc *p,
266    struct thread **newtdp, int flags, int pages, const char *tdname)
267{
268	struct thread *newtd, *oldtd;
269
270	if (!proc0.p_stats)
271		panic("kthread_add called too soon");
272
273	/* If no process supplied, put it on proc0 */
274	if (p == NULL)
275		p = &proc0;
276
277	/* Initialize our new td  */
278	newtd = thread_alloc(pages);
279	if (newtd == NULL)
280		return (ENOMEM);
281
282	PROC_LOCK(p);
283	if (p->p_state == PRS_ZOMBIE || (p->p_flag2 & P2_WEXIT) != 0) {
284		PROC_UNLOCK(p);
285		return (ESRCH);
286	}
287	oldtd = FIRST_THREAD_IN_PROC(p);
288
289	/*
290	 * Set the new thread pointer before the thread starts running: *newtdp
291	 * could be a pointer that is referenced by "func".
292	 */
293	if (newtdp != NULL)
294		*newtdp = newtd;
295
296	bzero(&newtd->td_startzero,
297	    __rangeof(struct thread, td_startzero, td_endzero));
298	bcopy(&oldtd->td_startcopy, &newtd->td_startcopy,
299	    __rangeof(struct thread, td_startcopy, td_endcopy));
300
301	/* set up arg0 for 'ps', et al */
302	strcpy(newtd->td_name, tdname);
303
304	TSTHREAD(newtd, newtd->td_name);
305
306	newtd->td_proc = p;  /* needed for cpu_copy_thread */
307	newtd->td_pflags |= TDP_KTHREAD;
308
309	/* might be further optimized for kthread */
310	cpu_copy_thread(newtd, oldtd);
311
312	/* put the designated function(arg) as the resume context */
313	cpu_fork_kthread_handler(newtd, func, arg);
314
315	thread_cow_get_proc(newtd, p);
316
317	/* This code is similar to thread_create() in kern_thr.c. */
318	p->p_flag |= P_HADTHREADS;
319	thread_link(newtd, p);
320	thread_lock(oldtd);
321	/* let the scheduler know about these things. */
322	sched_fork_thread(oldtd, newtd);
323	TD_SET_CAN_RUN(newtd);
324	thread_unlock(oldtd);
325	PROC_UNLOCK(p);
326
327	tidhash_add(newtd);
328
329	/* Avoid inheriting affinity from a random parent. */
330	cpuset_kernthread(newtd);
331#ifdef HWPMC_HOOKS
332	if (PMC_SYSTEM_SAMPLING_ACTIVE())
333		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_CREATE_LOG, NULL);
334#endif
335	/* Delay putting it on the run queue until now. */
336	if ((flags & RFSTOPPED) == 0) {
337		thread_lock(newtd);
338		sched_add(newtd, SRQ_BORING);
339	}
340	return (0);
341}
342
343int
344kthread_add(void (*func)(void *), void *arg, struct proc *p,
345    struct thread **newtdp, int flags, int pages, const char *fmt, ...)
346{
347	va_list ap;
348	int error;
349	char tdname[MAXCOMLEN + 1];
350
351	va_start(ap, fmt);
352	vsnprintf(tdname, sizeof(tdname), fmt, ap);
353	va_end(ap);
354	DROP_GIANT();
355	error = kthread_add1(func, arg, p, newtdp, flags, pages, tdname);
356	PICKUP_GIANT();
357	return (error);
358}
359
360void
361kthread_exit(void)
362{
363	struct proc *p;
364	struct thread *td;
365
366	td = curthread;
367	p = td->td_proc;
368
369#ifdef HWPMC_HOOKS
370	if (PMC_SYSTEM_SAMPLING_ACTIVE())
371		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT_LOG, NULL);
372#endif
373	/* A module may be waiting for us to exit. */
374	wakeup(td);
375
376	/*
377	 * The last exiting thread in a kernel process must tear down
378	 * the whole process.
379	 */
380	PROC_LOCK(p);
381	if (p->p_numthreads == 1) {
382		PROC_UNLOCK(p);
383		kproc_exit(0);
384	}
385
386	if (p->p_sysent->sv_ontdexit != NULL)
387		p->p_sysent->sv_ontdexit(td);
388
389	tidhash_remove(td);
390	umtx_thread_exit(td);
391	tdsigcleanup(td);
392	PROC_SLOCK(p);
393	thread_exit();
394}
395
396/*
397 * Advise a kernel process to suspend (or resume) in its main loop.
398 * Participation is voluntary.
399 */
400int
401kthread_suspend(struct thread *td, int timo)
402{
403	struct proc *p;
404
405	p = td->td_proc;
406
407	/*
408	 * td_pflags should not be read by any thread other than
409	 * curthread, but as long as this flag is invariant during the
410	 * thread's lifetime, it is OK to check its state.
411	 */
412	if ((td->td_pflags & TDP_KTHREAD) == 0)
413		return (EINVAL);
414
415	/*
416	 * The caller of the primitive should have already checked that the
417	 * thread is up and running, thus not being blocked by other
418	 * conditions.
419	 */
420	PROC_LOCK(p);
421	thread_lock(td);
422	td->td_flags |= TDF_KTH_SUSP;
423	thread_unlock(td);
424	return (msleep(&td->td_flags, &p->p_mtx, PPAUSE | PDROP, "suspkt",
425	    timo));
426}
427
428/*
429 * Resume a thread previously put asleep with kthread_suspend().
430 */
431int
432kthread_resume(struct thread *td)
433{
434	struct proc *p;
435
436	p = td->td_proc;
437
438	/*
439	 * td_pflags should not be read by any thread other than
440	 * curthread, but as long as this flag is invariant during the
441	 * thread's lifetime, it is OK to check its state.
442	 */
443	if ((td->td_pflags & TDP_KTHREAD) == 0)
444		return (EINVAL);
445
446	PROC_LOCK(p);
447	thread_lock(td);
448	td->td_flags &= ~TDF_KTH_SUSP;
449	thread_unlock(td);
450	wakeup(&td->td_flags);
451	PROC_UNLOCK(p);
452	return (0);
453}
454
455/*
456 * Used by the thread to poll as to whether it should yield/sleep
457 * and notify the caller that is has happened.
458 */
459void
460kthread_suspend_check(void)
461{
462	struct proc *p;
463	struct thread *td;
464
465	td = curthread;
466	p = td->td_proc;
467
468	if ((td->td_pflags & TDP_KTHREAD) == 0)
469		panic("%s: curthread is not a valid kthread", __func__);
470
471	/*
472	 * Setting the TDF_KTH_SUSP flag is protected by process lock.
473	 *
474	 * Do an unlocked read first to avoid serializing with all other threads
475	 * in the common case of not suspending.
476	 */
477	if ((td->td_flags & TDF_KTH_SUSP) == 0)
478		return;
479	PROC_LOCK(p);
480	while ((td->td_flags & TDF_KTH_SUSP) != 0) {
481		wakeup(&td->td_flags);
482		msleep(&td->td_flags, &p->p_mtx, PPAUSE, "ktsusp", 0);
483	}
484	PROC_UNLOCK(p);
485}
486
487int
488kproc_kthread_add(void (*func)(void *), void *arg,
489    struct proc **procptr, struct thread **tdptr,
490    int flags, int pages, const char *procname, const char *fmt, ...)
491{
492	int error;
493	va_list ap;
494	char buf[100];
495	struct thread *td;
496
497	if (*procptr == NULL) {
498		/*
499		 * Use RFSTOPPED to ensure that *tdptr is initialized before the
500		 * thread starts running.
501		 */
502		error = kproc_create(func, arg,
503		    procptr, flags | RFSTOPPED, pages, "%s", procname);
504		if (error)
505			return (error);
506		td = FIRST_THREAD_IN_PROC(*procptr);
507		if (tdptr)
508			*tdptr = td;
509		if ((flags & RFSTOPPED) == 0) {
510			thread_lock(td);
511			sched_add(td, SRQ_BORING);
512		}
513		va_start(ap, fmt);
514		vsnprintf(td->td_name, sizeof(td->td_name), fmt, ap);
515		va_end(ap);
516#ifdef KTR
517		sched_clear_tdname(td);
518#endif
519		return (0);
520	}
521	va_start(ap, fmt);
522	vsnprintf(buf, sizeof(buf), fmt, ap);
523	va_end(ap);
524	error = kthread_add(func, arg, *procptr,
525	    tdptr, flags, pages, "%s", buf);
526	return (error);
527}
528