lwproc.c revision 1.28 
1/*      $NetBSD: lwproc.c,v 1.28 2014/03/16 15:30:05 pooka Exp $	*/
2
3/*
4 * Copyright (c) 2010, 2011 Antti Kantee.  All Rights Reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following 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
16 * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
17 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
18 * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
21 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28#define RUMP__CURLWP_PRIVATE
29
30#include <sys/cdefs.h>
31__KERNEL_RCSID(0, "$NetBSD: lwproc.c,v 1.28 2014/03/16 15:30:05 pooka Exp $");
32
33#include <sys/param.h>
34#include <sys/atomic.h>
35#include <sys/filedesc.h>
36#include <sys/kauth.h>
37#include <sys/kmem.h>
38#include <sys/lwp.h>
39#include <sys/ktrace.h>
40#include <sys/pool.h>
41#include <sys/proc.h>
42#include <sys/queue.h>
43#include <sys/resourcevar.h>
44#include <sys/uidinfo.h>
45
46#include <rump/rumpuser.h>
47#include "rump_private.h"
48#include "rump_curlwp.h"
49
50struct emul *emul_default = &emul_netbsd;
51
52void
53rump_lwproc_init(void)
54{
55
56	lwproc_curlwpop(RUMPUSER_LWP_CREATE, &lwp0);
57}
58
59struct lwp *
60rump_lwproc_curlwp_hypercall(void)
61{
62
63	return rumpuser_curlwp();
64}
65
66void
67rump_lwproc_curlwp_set(struct lwp *l)
68{
69
70	KASSERT(curlwp == NULL);
71	lwproc_curlwpop(RUMPUSER_LWP_SET, l);
72}
73
74void
75rump_lwproc_curlwp_clear(struct lwp *l)
76{
77
78	KASSERT(l == curlwp);
79	lwproc_curlwpop(RUMPUSER_LWP_CLEAR, l);
80}
81
82static void
83lwproc_proc_free(struct proc *p)
84{
85	kauth_cred_t cred;
86
87	KASSERT(p->p_stat == SDYING || p->p_stat == SDEAD);
88
89#ifdef KTRACE
90	if (p->p_tracep) {
91		mutex_enter(&ktrace_lock);
92		ktrderef(p);
93		mutex_exit(&ktrace_lock);
94	}
95#endif
96
97	mutex_enter(proc_lock);
98
99	KASSERT(p->p_nlwps == 0);
100	KASSERT(LIST_EMPTY(&p->p_lwps));
101
102	LIST_REMOVE(p, p_list);
103	LIST_REMOVE(p, p_sibling);
104	proc_free_pid(p->p_pid); /* decrements nprocs */
105	proc_leavepgrp(p); /* releases proc_lock */
106
107	cred = p->p_cred;
108	chgproccnt(kauth_cred_getuid(cred), -1);
109	if (rump_proc_vfs_release)
110		rump_proc_vfs_release(p);
111
112	lim_free(p->p_limit);
113	pstatsfree(p->p_stats);
114	kauth_cred_free(p->p_cred);
115	proc_finispecific(p);
116
117	mutex_obj_free(p->p_lock);
118	mutex_destroy(&p->p_stmutex);
119	mutex_destroy(&p->p_auxlock);
120	rw_destroy(&p->p_reflock);
121	cv_destroy(&p->p_waitcv);
122	cv_destroy(&p->p_lwpcv);
123
124	/* non-kernel vmspaces are not shared */
125	if (!RUMP_LOCALPROC_P(p)) {
126		KASSERT(p->p_vmspace->vm_refcnt == 1);
127		kmem_free(p->p_vmspace, sizeof(*p->p_vmspace));
128	}
129
130	proc_free_mem(p);
131}
132
133/*
134 * Allocate a new process.  Mostly mimic fork by
135 * copying the properties of the parent.  However, there are some
136 * differences.
137 *
138 * Switch to the new lwp and return a pointer to it.
139 */
140static struct proc *
141lwproc_newproc(struct proc *parent, int flags)
142{
143	uid_t uid = kauth_cred_getuid(parent->p_cred);
144	struct proc *p;
145
146	/* maxproc not enforced */
147	atomic_inc_uint(&nprocs);
148
149	/* allocate process */
150	p = proc_alloc();
151	memset(&p->p_startzero, 0,
152	    offsetof(struct proc, p_endzero)
153	      - offsetof(struct proc, p_startzero));
154	memcpy(&p->p_startcopy, &parent->p_startcopy,
155	    offsetof(struct proc, p_endcopy)
156	      - offsetof(struct proc, p_startcopy));
157
158	/* some other garbage we need to zero */
159	p->p_sigacts = NULL;
160	p->p_aio = NULL;
161	p->p_dtrace = NULL;
162	p->p_mqueue_cnt = p->p_exitsig = 0;
163	p->p_flag = p->p_sflag = p->p_slflag = p->p_lflag = p->p_stflag = 0;
164	p->p_trace_enabled = 0;
165	p->p_xstat = p->p_acflag = 0;
166	p->p_stackbase = 0;
167
168	p->p_stats = pstatscopy(parent->p_stats);
169
170	p->p_vmspace = vmspace_kernel();
171	p->p_emul = emul_default;
172#ifdef __HAVE_SYSCALL_INTERN
173	p->p_emul->e_syscall_intern(p);
174#endif
175	if (*parent->p_comm)
176		strcpy(p->p_comm, parent->p_comm);
177	else
178		strcpy(p->p_comm, "rumproc");
179
180	if ((flags & RUMP_RFCFDG) == 0)
181		KASSERT(parent == curproc);
182	if (flags & RUMP_RFFDG)
183		p->p_fd = fd_copy();
184	else if (flags & RUMP_RFCFDG)
185		p->p_fd = fd_init(NULL);
186	else
187		fd_share(p);
188
189	lim_addref(parent->p_limit);
190	p->p_limit = parent->p_limit;
191
192	LIST_INIT(&p->p_lwps);
193	LIST_INIT(&p->p_children);
194
195	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
196	mutex_init(&p->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
197	mutex_init(&p->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
198	rw_init(&p->p_reflock);
199	cv_init(&p->p_waitcv, "pwait");
200	cv_init(&p->p_lwpcv, "plwp");
201
202	p->p_pptr = parent;
203	p->p_ppid = parent->p_pid;
204	p->p_stat = SACTIVE;
205
206	kauth_proc_fork(parent, p);
207
208	/* initialize cwd in rump kernels with vfs */
209	if (rump_proc_vfs_init)
210		rump_proc_vfs_init(p);
211
212	chgproccnt(uid, 1); /* not enforced */
213
214	/* publish proc various proc lists */
215	mutex_enter(proc_lock);
216	LIST_INSERT_HEAD(&allproc, p, p_list);
217	LIST_INSERT_HEAD(&parent->p_children, p, p_sibling);
218	LIST_INSERT_AFTER(parent, p, p_pglist);
219	mutex_exit(proc_lock);
220
221	return p;
222}
223
224static void
225lwproc_freelwp(struct lwp *l)
226{
227	struct proc *p;
228
229	p = l->l_proc;
230	mutex_enter(p->p_lock);
231
232	KASSERT(l->l_flag & LW_WEXIT);
233	KASSERT(l->l_refcnt == 0);
234
235	/* ok, zero references, continue with nuke */
236	LIST_REMOVE(l, l_sibling);
237	KASSERT(p->p_nlwps >= 1);
238	if (--p->p_nlwps == 0) {
239		KASSERT(p != &proc0);
240		p->p_stat = SDEAD;
241	}
242	cv_broadcast(&p->p_lwpcv); /* nobody sleeps on this in a rump kernel? */
243	kauth_cred_free(l->l_cred);
244	mutex_exit(p->p_lock);
245
246	mutex_enter(proc_lock);
247	LIST_REMOVE(l, l_list);
248	mutex_exit(proc_lock);
249
250	if (l->l_name)
251		kmem_free(l->l_name, MAXCOMLEN);
252	lwp_finispecific(l);
253
254	lwproc_curlwpop(RUMPUSER_LWP_DESTROY, l);
255	membar_exit();
256	kmem_free(l, sizeof(*l));
257
258	if (p->p_stat == SDEAD)
259		lwproc_proc_free(p);
260}
261
262extern kmutex_t unruntime_lock;
263
264/*
265 * called with p_lock held, releases lock before return
266 */
267static void
268lwproc_makelwp(struct proc *p, struct lwp *l, bool doswitch, bool procmake)
269{
270
271	p->p_nlwps++;
272	l->l_refcnt = 1;
273	l->l_proc = p;
274
275	l->l_lid = p->p_nlwpid++;
276	LIST_INSERT_HEAD(&p->p_lwps, l, l_sibling);
277
278	l->l_fd = p->p_fd;
279	l->l_cpu = rump_cpu;
280	l->l_target_cpu = rump_cpu; /* Initial target CPU always the same */
281	l->l_stat = LSRUN;
282	l->l_mutex = &unruntime_lock;
283	TAILQ_INIT(&l->l_ld_locks);
284	mutex_exit(p->p_lock);
285
286	lwp_update_creds(l);
287	lwp_initspecific(l);
288
289	membar_enter();
290	lwproc_curlwpop(RUMPUSER_LWP_CREATE, l);
291	if (doswitch) {
292		rump_lwproc_switch(l);
293	}
294
295	/* filedesc already has refcount 1 when process is created */
296	if (!procmake) {
297		fd_hold(l);
298	}
299
300	mutex_enter(proc_lock);
301	LIST_INSERT_HEAD(&alllwp, l, l_list);
302	mutex_exit(proc_lock);
303}
304
305struct lwp *
306rump__lwproc_alloclwp(struct proc *p)
307{
308	struct lwp *l;
309	bool newproc = false;
310
311	if (p == NULL) {
312		p = lwproc_newproc(&proc0, 0);
313		newproc = true;
314	}
315
316	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
317
318	mutex_enter(p->p_lock);
319	KASSERT((p->p_sflag & PS_RUMP_LWPEXIT) == 0);
320	lwproc_makelwp(p, l, false, newproc);
321
322	return l;
323}
324
325int
326rump_lwproc_newlwp(pid_t pid)
327{
328	struct proc *p;
329	struct lwp *l;
330
331	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
332	mutex_enter(proc_lock);
333	p = proc_find_raw(pid);
334	if (p == NULL) {
335		mutex_exit(proc_lock);
336		kmem_free(l, sizeof(*l));
337		return ESRCH;
338	}
339	mutex_enter(p->p_lock);
340	if (p->p_sflag & PS_RUMP_LWPEXIT) {
341		mutex_exit(proc_lock);
342		mutex_exit(p->p_lock);
343		kmem_free(l, sizeof(*l));
344		return EBUSY;
345	}
346	mutex_exit(proc_lock);
347	lwproc_makelwp(p, l, true, false);
348
349	return 0;
350}
351
352int
353rump_lwproc_rfork(int flags)
354{
355	struct proc *p;
356	struct lwp *l;
357
358	if (flags & ~(RUMP_RFFDG|RUMP_RFCFDG) ||
359	    (~flags & (RUMP_RFFDG|RUMP_RFCFDG)) == 0)
360		return EINVAL;
361
362	p = lwproc_newproc(curproc, flags);
363	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
364	mutex_enter(p->p_lock);
365	KASSERT((p->p_sflag & PS_RUMP_LWPEXIT) == 0);
366	lwproc_makelwp(p, l, true, true);
367
368	return 0;
369}
370
371/*
372 * Switch to a new process/thread.  Release previous one if
373 * deemed to be exiting.  This is considered a slow path for
374 * rump kernel entry.
375 */
376void
377rump_lwproc_switch(struct lwp *newlwp)
378{
379	struct lwp *l = curlwp;
380
381	KASSERT(!(l->l_flag & LW_WEXIT) || newlwp);
382
383	if (__predict_false(newlwp && (newlwp->l_pflag & LP_RUNNING)))
384		panic("lwp %p (%d:%d) already running",
385		    newlwp, newlwp->l_proc->p_pid, newlwp->l_lid);
386
387	if (newlwp == NULL) {
388		l->l_pflag &= ~LP_RUNNING;
389		l->l_flag |= LW_RUMP_CLEAR;
390		return;
391	}
392
393	/* fd_free() must be called from curlwp context.  talk about ugh */
394	if (l->l_flag & LW_WEXIT) {
395		fd_free();
396	}
397
398	KERNEL_UNLOCK_ALL(NULL, &l->l_biglocks);
399	lwproc_curlwpop(RUMPUSER_LWP_CLEAR, l);
400
401	newlwp->l_cpu = newlwp->l_target_cpu = l->l_cpu;
402	newlwp->l_mutex = l->l_mutex;
403	newlwp->l_pflag |= LP_RUNNING;
404
405	lwproc_curlwpop(RUMPUSER_LWP_SET, newlwp);
406	curcpu()->ci_curlwp = newlwp;
407	KERNEL_LOCK(newlwp->l_biglocks, NULL);
408
409	/*
410	 * Check if the thread should get a signal.  This is
411	 * mostly to satisfy the "record" rump sigmodel.
412	 */
413	mutex_enter(newlwp->l_proc->p_lock);
414	if (sigispending(newlwp, 0)) {
415		newlwp->l_flag |= LW_PENDSIG;
416	}
417	mutex_exit(newlwp->l_proc->p_lock);
418
419	l->l_mutex = &unruntime_lock;
420	l->l_pflag &= ~LP_RUNNING;
421	l->l_flag &= ~LW_PENDSIG;
422	l->l_stat = LSRUN;
423
424	if (l->l_flag & LW_WEXIT) {
425		lwproc_freelwp(l);
426	}
427}
428
429/*
430 * Mark the current thread to be released upon return from
431 * kernel.
432 */
433void
434rump_lwproc_releaselwp(void)
435{
436	struct lwp *l = curlwp;
437
438	if (l->l_refcnt == 0 || l->l_flag & LW_WEXIT)
439		panic("releasing non-pertinent lwp");
440
441	rump__lwproc_lwprele();
442	KASSERT(l->l_refcnt == 0 && (l->l_flag & LW_WEXIT));
443}
444
445/*
446 * In-kernel routines used to add and remove references for the
447 * current thread.  The main purpose is to make it possible for
448 * implicit threads to persist over scheduling operations in
449 * rump kernel drivers.  Note that we don't need p_lock in a
450 * rump kernel, since we do refcounting only for curlwp.
451 */
452void
453rump__lwproc_lwphold(void)
454{
455	struct lwp *l = curlwp;
456
457	l->l_refcnt++;
458	l->l_flag &= ~LW_WEXIT;
459}
460
461void
462rump__lwproc_lwprele(void)
463{
464	struct lwp *l = curlwp;
465
466	l->l_refcnt--;
467	if (l->l_refcnt == 0)
468		l->l_flag |= LW_WEXIT;
469}
470
471struct lwp *
472rump_lwproc_curlwp(void)
473{
474	struct lwp *l = curlwp;
475
476	if (l->l_flag & LW_WEXIT)
477		return NULL;
478	return l;
479}
480
481/* this interface is under construction (like the proverbial 90's web page) */
482int rump_i_know_what_i_am_doing_with_sysents = 0;
483void
484rump_lwproc_sysent_usenative()
485{
486
487	if (!rump_i_know_what_i_am_doing_with_sysents)
488		panic("don't use rump_lwproc_sysent_usenative()");
489	curproc->p_emul = &emul_netbsd;
490}
491