xref: /freebsd/sys/kern/kern_kthread.c (revision 8524dc53fd4c6b79d75b82cb82f3ac72fc25e85f)
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/cdefs.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/cpuset.h>
33 #include <sys/kthread.h>
34 #include <sys/lock.h>
35 #include <sys/mutex.h>
36 #include <sys/proc.h>
37 #include <sys/resourcevar.h>
38 #include <sys/rwlock.h>
39 #include <sys/signalvar.h>
40 #include <sys/sysent.h>
41 #include <sys/sx.h>
42 #include <sys/umtxvar.h>
43 #include <sys/unistd.h>
44 #include <sys/wait.h>
45 #include <sys/sched.h>
46 #include <sys/tslog.h>
47 #include <vm/vm.h>
48 #include <vm/vm_extern.h>
49 
50 #include <machine/stdarg.h>
51 
52 /*
53  * Start a kernel process.  This is called after a fork() call in
54  * mi_startup() in the file kern/init_main.c.
55  *
56  * This function is used to start "internal" daemons and intended
57  * to be called from SYSINIT().
58  */
59 void
60 kproc_start(const void *udata)
61 {
62 	const struct kproc_desc	*kp = udata;
63 	int error;
64 
65 	error = kproc_create((void (*)(void *))kp->func, NULL,
66 	    kp->global_procpp, 0, 0, "%s", kp->arg0);
67 	if (error)
68 		panic("kproc_start: %s: error %d", kp->arg0, error);
69 }
70 
71 /*
72  * Create a kernel process/thread/whatever.  It shares its address space
73  * with proc0 - ie: kernel only.
74  *
75  * func is the function to start.
76  * arg is the parameter to pass to function on first startup.
77  * newpp is the return value pointing to the thread's struct proc.
78  * flags are flags to fork1 (in unistd.h)
79  * fmt and following will be *printf'd into (*newpp)->p_comm (for ps, etc.).
80  */
81 static int
82 kproc_create1(void (*func)(void *), void *arg,
83     struct proc **newpp, int flags, int pages, const char *tdname)
84 {
85 	struct fork_req fr;
86 	int error;
87 	struct thread *td;
88 	struct proc *p2;
89 
90 	if (!proc0.p_stats)
91 		panic("kproc_create called too soon");
92 
93 	bzero(&fr, sizeof(fr));
94 	fr.fr_flags = RFMEM | RFFDG | RFPROC | RFSTOPPED | flags;
95 	fr.fr_flags2 = FR2_KPROC;
96 	fr.fr_pages = pages;
97 	fr.fr_procp = &p2;
98 	error = fork1(&thread0, &fr);
99 	if (error != 0)
100 		return (error);
101 
102 	/* save a global descriptor, if desired */
103 	if (newpp != NULL)
104 		*newpp = p2;
105 
106 	/* set up arg0 for 'ps', et al */
107 	strcpy(p2->p_comm, tdname);
108 	td = FIRST_THREAD_IN_PROC(p2);
109 	strcpy(td->td_name, tdname);
110 #ifdef KTR
111 	sched_clear_tdname(td);
112 #endif
113 	TSTHREAD(td, td->td_name);
114 #ifdef HWPMC_HOOKS
115 	if (PMC_SYSTEM_SAMPLING_ACTIVE()) {
116 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_PROC_CREATE_LOG, p2);
117 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_CREATE_LOG, NULL);
118 	}
119 #endif
120 
121 	/* call the processes' main()... */
122 	cpu_fork_kthread_handler(td, func, arg);
123 
124 	/* Avoid inheriting affinity from a random parent. */
125 	cpuset_kernthread(td);
126 	thread_lock(td);
127 	TD_SET_CAN_RUN(td);
128 	sched_prio(td, PVM);
129 	sched_user_prio(td, PUSER);
130 
131 	/* Delay putting it on the run queue until now. */
132 	if ((flags & RFSTOPPED) == 0)
133 		sched_add(td, SRQ_BORING);
134 	else
135 		thread_unlock(td);
136 
137 	return (0);
138 }
139 
140 int
141 kproc_create(void (*func)(void *), void *arg,
142     struct proc **newpp, int flags, int pages, const char *fmt, ...)
143 {
144 	va_list ap;
145 	int error;
146 	char tdname[MAXCOMLEN + 1];
147 
148 	va_start(ap, fmt);
149 	vsnprintf(tdname, sizeof(tdname), fmt, ap);
150 	va_end(ap);
151 	DROP_GIANT();
152 	error = kproc_create1(func, arg, newpp, flags, pages, tdname);
153 	PICKUP_GIANT();
154 	return (error);
155 }
156 
157 void
158 kproc_exit(int ecode)
159 {
160 	struct thread *td;
161 	struct proc *p;
162 
163 	td = curthread;
164 	p = td->td_proc;
165 
166 	/*
167 	 * Reparent curthread from proc0 to init so that the zombie
168 	 * is harvested.
169 	 */
170 	sx_xlock(&proctree_lock);
171 	PROC_LOCK(p);
172 	proc_reparent(p, initproc, true);
173 	PROC_UNLOCK(p);
174 	sx_xunlock(&proctree_lock);
175 
176 	/*
177 	 * Wakeup anyone waiting for us to exit.
178 	 */
179 	wakeup(p);
180 
181 	/* Buh-bye! */
182 	exit1(td, ecode, 0);
183 }
184 
185 /*
186  * Advise a kernel process to suspend (or resume) in its main loop.
187  * Participation is voluntary.
188  */
189 int
190 kproc_suspend(struct proc *p, int timo)
191 {
192 	/*
193 	 * Make sure this is indeed a system process and we can safely
194 	 * use the p_siglist field.
195 	 */
196 	PROC_LOCK(p);
197 	if ((p->p_flag & P_KPROC) == 0) {
198 		PROC_UNLOCK(p);
199 		return (EINVAL);
200 	}
201 	SIGADDSET(p->p_siglist, SIGSTOP);
202 	wakeup(p);
203 	return (msleep(&p->p_siglist, &p->p_mtx, PPAUSE | PDROP,
204 	    "suspkp", timo));
205 }
206 
207 int
208 kproc_resume(struct proc *p)
209 {
210 	/*
211 	 * Make sure this is indeed a system process and we can safely
212 	 * use the p_siglist field.
213 	 */
214 	PROC_LOCK(p);
215 	if ((p->p_flag & P_KPROC) == 0) {
216 		PROC_UNLOCK(p);
217 		return (EINVAL);
218 	}
219 	SIGDELSET(p->p_siglist, SIGSTOP);
220 	PROC_UNLOCK(p);
221 	wakeup(&p->p_siglist);
222 	return (0);
223 }
224 
225 void
226 kproc_suspend_check(struct proc *p)
227 {
228 	PROC_LOCK(p);
229 	while (SIGISMEMBER(p->p_siglist, SIGSTOP)) {
230 		wakeup(&p->p_siglist);
231 		msleep(&p->p_siglist, &p->p_mtx, PPAUSE, "kpsusp", 0);
232 	}
233 	PROC_UNLOCK(p);
234 }
235 
236 /*
237  * Start a kernel thread.
238  *
239  * This function is used to start "internal" daemons and intended
240  * to be called from SYSINIT().
241  */
242 
243 void
244 kthread_start(const void *udata)
245 {
246 	const struct kthread_desc *kp = udata;
247 	int error;
248 
249 	error = kthread_add((void (*)(void *))kp->func, NULL,
250 	    NULL, kp->global_threadpp, 0, 0, "%s", kp->arg0);
251 	if (error)
252 		panic("kthread_start: %s: error %d", kp->arg0, error);
253 }
254 
255 /*
256  * Create a kernel thread.  It shares its address space
257  * with proc0 - ie: kernel only.
258  *
259  * func is the function to start.
260  * arg is the parameter to pass to function on first startup.
261  * newtdp is the return value pointing to the thread's struct thread.
262  *  ** XXX fix this --> flags are flags to fork1 (in unistd.h)
263  * fmt and following will be *printf'd into (*newtd)->td_name (for ps, etc.).
264  */
265 static int
266 kthread_add1(void (*func)(void *), void *arg, struct proc *p,
267     struct thread **newtdp, int flags, int pages, const char *tdname)
268 {
269 	struct thread *newtd, *oldtd;
270 
271 	if (!proc0.p_stats)
272 		panic("kthread_add called too soon");
273 
274 	/* If no process supplied, put it on proc0 */
275 	if (p == NULL)
276 		p = &proc0;
277 
278 	/* Initialize our new td  */
279 	newtd = thread_alloc(pages);
280 	if (newtd == NULL)
281 		return (ENOMEM);
282 
283 	PROC_LOCK(p);
284 	oldtd = FIRST_THREAD_IN_PROC(p);
285 
286 	bzero(&newtd->td_startzero,
287 	    __rangeof(struct thread, td_startzero, td_endzero));
288 	bcopy(&oldtd->td_startcopy, &newtd->td_startcopy,
289 	    __rangeof(struct thread, td_startcopy, td_endcopy));
290 
291 	/* set up arg0 for 'ps', et al */
292 	strcpy(newtd->td_name, tdname);
293 
294 	TSTHREAD(newtd, newtd->td_name);
295 
296 	newtd->td_proc = p;  /* needed for cpu_copy_thread */
297 	newtd->td_pflags |= TDP_KTHREAD;
298 
299 	/* might be further optimized for kthread */
300 	cpu_copy_thread(newtd, oldtd);
301 
302 	/* put the designated function(arg) as the resume context */
303 	cpu_fork_kthread_handler(newtd, func, arg);
304 
305 	thread_cow_get_proc(newtd, p);
306 
307 	/* This code is similar to thread_create() in kern_thr.c. */
308 	p->p_flag |= P_HADTHREADS;
309 	thread_link(newtd, p);
310 	thread_lock(oldtd);
311 	/* let the scheduler know about these things. */
312 	sched_fork_thread(oldtd, newtd);
313 	TD_SET_CAN_RUN(newtd);
314 	thread_unlock(oldtd);
315 	PROC_UNLOCK(p);
316 
317 	tidhash_add(newtd);
318 
319 	/* Avoid inheriting affinity from a random parent. */
320 	cpuset_kernthread(newtd);
321 #ifdef HWPMC_HOOKS
322 	if (PMC_SYSTEM_SAMPLING_ACTIVE())
323 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_CREATE_LOG, NULL);
324 #endif
325 	/* Delay putting it on the run queue until now. */
326 	if ((flags & RFSTOPPED) == 0) {
327 		thread_lock(newtd);
328 		sched_add(newtd, SRQ_BORING);
329 	}
330 	if (newtdp)
331 		*newtdp = newtd;
332 	return (0);
333 }
334 
335 int
336 kthread_add(void (*func)(void *), void *arg, struct proc *p,
337     struct thread **newtdp, int flags, int pages, const char *fmt, ...)
338 {
339 	va_list ap;
340 	int error;
341 	char tdname[MAXCOMLEN + 1];
342 
343 	va_start(ap, fmt);
344 	vsnprintf(tdname, sizeof(tdname), fmt, ap);
345 	va_end(ap);
346 	DROP_GIANT();
347 	error = kthread_add1(func, arg, p, newtdp, flags, pages, tdname);
348 	PICKUP_GIANT();
349 	return (error);
350 }
351 
352 void
353 kthread_exit(void)
354 {
355 	struct proc *p;
356 	struct thread *td;
357 
358 	td = curthread;
359 	p = td->td_proc;
360 
361 #ifdef HWPMC_HOOKS
362 	if (PMC_SYSTEM_SAMPLING_ACTIVE())
363 		PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_THR_EXIT_LOG, NULL);
364 #endif
365 	/* A module may be waiting for us to exit. */
366 	wakeup(td);
367 
368 	/*
369 	 * The last exiting thread in a kernel process must tear down
370 	 * the whole process.
371 	 */
372 	PROC_LOCK(p);
373 	if (p->p_numthreads == 1) {
374 		PROC_UNLOCK(p);
375 		kproc_exit(0);
376 	}
377 
378 	if (p->p_sysent->sv_ontdexit != NULL)
379 		p->p_sysent->sv_ontdexit(td);
380 
381 	tidhash_remove(td);
382 	umtx_thread_exit(td);
383 	tdsigcleanup(td);
384 	PROC_SLOCK(p);
385 	thread_exit();
386 }
387 
388 /*
389  * Advise a kernel process to suspend (or resume) in its main loop.
390  * Participation is voluntary.
391  */
392 int
393 kthread_suspend(struct thread *td, int timo)
394 {
395 	struct proc *p;
396 
397 	p = td->td_proc;
398 
399 	/*
400 	 * td_pflags should not be read by any thread other than
401 	 * curthread, but as long as this flag is invariant during the
402 	 * thread's lifetime, it is OK to check its state.
403 	 */
404 	if ((td->td_pflags & TDP_KTHREAD) == 0)
405 		return (EINVAL);
406 
407 	/*
408 	 * The caller of the primitive should have already checked that the
409 	 * thread is up and running, thus not being blocked by other
410 	 * conditions.
411 	 */
412 	PROC_LOCK(p);
413 	thread_lock(td);
414 	td->td_flags |= TDF_KTH_SUSP;
415 	thread_unlock(td);
416 	return (msleep(&td->td_flags, &p->p_mtx, PPAUSE | PDROP, "suspkt",
417 	    timo));
418 }
419 
420 /*
421  * Resume a thread previously put asleep with kthread_suspend().
422  */
423 int
424 kthread_resume(struct thread *td)
425 {
426 	struct proc *p;
427 
428 	p = td->td_proc;
429 
430 	/*
431 	 * td_pflags should not be read by any thread other than
432 	 * curthread, but as long as this flag is invariant during the
433 	 * thread's lifetime, it is OK to check its state.
434 	 */
435 	if ((td->td_pflags & TDP_KTHREAD) == 0)
436 		return (EINVAL);
437 
438 	PROC_LOCK(p);
439 	thread_lock(td);
440 	td->td_flags &= ~TDF_KTH_SUSP;
441 	thread_unlock(td);
442 	wakeup(&td->td_flags);
443 	PROC_UNLOCK(p);
444 	return (0);
445 }
446 
447 /*
448  * Used by the thread to poll as to whether it should yield/sleep
449  * and notify the caller that is has happened.
450  */
451 void
452 kthread_suspend_check(void)
453 {
454 	struct proc *p;
455 	struct thread *td;
456 
457 	td = curthread;
458 	p = td->td_proc;
459 
460 	if ((td->td_pflags & TDP_KTHREAD) == 0)
461 		panic("%s: curthread is not a valid kthread", __func__);
462 
463 	/*
464 	 * Setting the TDF_KTH_SUSP flag is protected by process lock.
465 	 *
466 	 * Do an unlocked read first to avoid serializing with all other threads
467 	 * in the common case of not suspending.
468 	 */
469 	if ((td->td_flags & TDF_KTH_SUSP) == 0)
470 		return;
471 	PROC_LOCK(p);
472 	while ((td->td_flags & TDF_KTH_SUSP) != 0) {
473 		wakeup(&td->td_flags);
474 		msleep(&td->td_flags, &p->p_mtx, PPAUSE, "ktsusp", 0);
475 	}
476 	PROC_UNLOCK(p);
477 }
478 
479 int
480 kproc_kthread_add(void (*func)(void *), void *arg,
481     struct proc **procptr, struct thread **tdptr,
482     int flags, int pages, const char *procname, const char *fmt, ...)
483 {
484 	int error;
485 	va_list ap;
486 	char buf[100];
487 	struct thread *td;
488 
489 	if (*procptr == NULL) {
490 		error = kproc_create(func, arg,
491 		    procptr, flags, pages, "%s", procname);
492 		if (error)
493 			return (error);
494 		td = FIRST_THREAD_IN_PROC(*procptr);
495 		if (tdptr)
496 			*tdptr = td;
497 		va_start(ap, fmt);
498 		vsnprintf(td->td_name, sizeof(td->td_name), fmt, ap);
499 		va_end(ap);
500 #ifdef KTR
501 		sched_clear_tdname(td);
502 #endif
503 		return (0);
504 	}
505 	va_start(ap, fmt);
506 	vsnprintf(buf, sizeof(buf), fmt, ap);
507 	va_end(ap);
508 	error = kthread_add(func, arg, *procptr,
509 	    tdptr, flags, pages, "%s", buf);
510 	return (error);
511 }
512