Home | History | Annotate | Line # | Download | only in rumpkern
lwproc.c revision 1.25
      1 /*      $NetBSD: lwproc.c,v 1.25 2013/12/09 16:21:15 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 #include <sys/cdefs.h>
     29 __KERNEL_RCSID(0, "$NetBSD: lwproc.c,v 1.25 2013/12/09 16:21:15 pooka Exp $");
     30 
     31 #include <sys/param.h>
     32 #include <sys/atomic.h>
     33 #include <sys/filedesc.h>
     34 #include <sys/kauth.h>
     35 #include <sys/kmem.h>
     36 #include <sys/lwp.h>
     37 #include <sys/ktrace.h>
     38 #include <sys/pool.h>
     39 #include <sys/proc.h>
     40 #include <sys/queue.h>
     41 #include <sys/resourcevar.h>
     42 #include <sys/uidinfo.h>
     43 
     44 #include <rump/rumpuser.h>
     45 
     46 #include "rump_private.h"
     47 
     48 struct emul *emul_default = &emul_netbsd;
     49 
     50 static void
     51 lwproc_proc_free(struct proc *p)
     52 {
     53 	kauth_cred_t cred;
     54 
     55 	KASSERT(p->p_stat == SDYING || p->p_stat == SDEAD);
     56 
     57 #ifdef KTRACE
     58 	if (p->p_tracep) {
     59 		mutex_enter(&ktrace_lock);
     60 		ktrderef(p);
     61 		mutex_exit(&ktrace_lock);
     62 	}
     63 #endif
     64 
     65 	mutex_enter(proc_lock);
     66 
     67 	KASSERT(p->p_nlwps == 0);
     68 	KASSERT(LIST_EMPTY(&p->p_lwps));
     69 
     70 	LIST_REMOVE(p, p_list);
     71 	LIST_REMOVE(p, p_sibling);
     72 	proc_free_pid(p->p_pid); /* decrements nprocs */
     73 	proc_leavepgrp(p); /* releases proc_lock */
     74 
     75 	cred = p->p_cred;
     76 	chgproccnt(kauth_cred_getuid(cred), -1);
     77 	if (rump_proc_vfs_release)
     78 		rump_proc_vfs_release(p);
     79 
     80 	lim_free(p->p_limit);
     81 	pstatsfree(p->p_stats);
     82 	kauth_cred_free(p->p_cred);
     83 	proc_finispecific(p);
     84 
     85 	mutex_obj_free(p->p_lock);
     86 	mutex_destroy(&p->p_stmutex);
     87 	mutex_destroy(&p->p_auxlock);
     88 	rw_destroy(&p->p_reflock);
     89 	cv_destroy(&p->p_waitcv);
     90 	cv_destroy(&p->p_lwpcv);
     91 
     92 	/* non-kernel vmspaces are not shared */
     93 	if (!RUMP_LOCALPROC_P(p)) {
     94 		KASSERT(p->p_vmspace->vm_refcnt == 1);
     95 		kmem_free(p->p_vmspace, sizeof(*p->p_vmspace));
     96 	}
     97 
     98 	proc_free_mem(p);
     99 }
    100 
    101 /*
    102  * Allocate a new process.  Mostly mimic fork by
    103  * copying the properties of the parent.  However, there are some
    104  * differences.
    105  *
    106  * Switch to the new lwp and return a pointer to it.
    107  */
    108 static struct proc *
    109 lwproc_newproc(struct proc *parent, int flags)
    110 {
    111 	uid_t uid = kauth_cred_getuid(parent->p_cred);
    112 	struct proc *p;
    113 
    114 	/* maxproc not enforced */
    115 	atomic_inc_uint(&nprocs);
    116 
    117 	/* allocate process */
    118 	p = proc_alloc();
    119 	memset(&p->p_startzero, 0,
    120 	    offsetof(struct proc, p_endzero)
    121 	      - offsetof(struct proc, p_startzero));
    122 	memcpy(&p->p_startcopy, &parent->p_startcopy,
    123 	    offsetof(struct proc, p_endcopy)
    124 	      - offsetof(struct proc, p_startcopy));
    125 
    126 	/* some other garbage we need to zero */
    127 	p->p_sigacts = NULL;
    128 	p->p_aio = NULL;
    129 	p->p_dtrace = NULL;
    130 	p->p_mqueue_cnt = p->p_exitsig = 0;
    131 	p->p_flag = p->p_sflag = p->p_slflag = p->p_lflag = p->p_stflag = 0;
    132 	p->p_trace_enabled = 0;
    133 	p->p_xstat = p->p_acflag = 0;
    134 	p->p_stackbase = 0;
    135 
    136 	p->p_stats = pstatscopy(parent->p_stats);
    137 
    138 	p->p_vmspace = vmspace_kernel();
    139 	p->p_emul = emul_default;
    140 	if (*parent->p_comm)
    141 		strcpy(p->p_comm, parent->p_comm);
    142 	else
    143 		strcpy(p->p_comm, "rumproc");
    144 
    145 	if ((flags & RUMP_RFCFDG) == 0)
    146 		KASSERT(parent == curproc);
    147 	if (flags & RUMP_RFFDG)
    148 		p->p_fd = fd_copy();
    149 	else if (flags & RUMP_RFCFDG)
    150 		p->p_fd = fd_init(NULL);
    151 	else
    152 		fd_share(p);
    153 
    154 	lim_addref(parent->p_limit);
    155 	p->p_limit = parent->p_limit;
    156 
    157 	LIST_INIT(&p->p_lwps);
    158 	LIST_INIT(&p->p_children);
    159 
    160 	p->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
    161 	mutex_init(&p->p_stmutex, MUTEX_DEFAULT, IPL_HIGH);
    162 	mutex_init(&p->p_auxlock, MUTEX_DEFAULT, IPL_NONE);
    163 	rw_init(&p->p_reflock);
    164 	cv_init(&p->p_waitcv, "pwait");
    165 	cv_init(&p->p_lwpcv, "plwp");
    166 
    167 	p->p_pptr = parent;
    168 	p->p_ppid = parent->p_pid;
    169 	p->p_stat = SACTIVE;
    170 
    171 	kauth_proc_fork(parent, p);
    172 
    173 	/* initialize cwd in rump kernels with vfs */
    174 	if (rump_proc_vfs_init)
    175 		rump_proc_vfs_init(p);
    176 
    177 	chgproccnt(uid, 1); /* not enforced */
    178 
    179 	/* publish proc various proc lists */
    180 	mutex_enter(proc_lock);
    181 	LIST_INSERT_HEAD(&allproc, p, p_list);
    182 	LIST_INSERT_HEAD(&parent->p_children, p, p_sibling);
    183 	LIST_INSERT_AFTER(parent, p, p_pglist);
    184 	mutex_exit(proc_lock);
    185 
    186 	return p;
    187 }
    188 
    189 static void
    190 lwproc_freelwp(struct lwp *l)
    191 {
    192 	struct proc *p;
    193 
    194 	p = l->l_proc;
    195 	mutex_enter(p->p_lock);
    196 
    197 	KASSERT(l->l_flag & LW_WEXIT);
    198 	KASSERT(l->l_refcnt == 0);
    199 
    200 	/* ok, zero references, continue with nuke */
    201 	LIST_REMOVE(l, l_sibling);
    202 	KASSERT(p->p_nlwps >= 1);
    203 	if (--p->p_nlwps == 0) {
    204 		KASSERT(p != &proc0);
    205 		p->p_stat = SDEAD;
    206 	}
    207 	cv_broadcast(&p->p_lwpcv); /* nobody sleeps on this in a rump kernel? */
    208 	kauth_cred_free(l->l_cred);
    209 	mutex_exit(p->p_lock);
    210 
    211 	mutex_enter(proc_lock);
    212 	LIST_REMOVE(l, l_list);
    213 	mutex_exit(proc_lock);
    214 
    215 	if (l->l_name)
    216 		kmem_free(l->l_name, MAXCOMLEN);
    217 	lwp_finispecific(l);
    218 
    219 	rumpuser_curlwpop(RUMPUSER_LWP_DESTROY, l);
    220 	membar_exit();
    221 	kmem_free(l, sizeof(*l));
    222 
    223 	if (p->p_stat == SDEAD)
    224 		lwproc_proc_free(p);
    225 }
    226 
    227 extern kmutex_t unruntime_lock;
    228 
    229 /*
    230  * called with p_lock held, releases lock before return
    231  */
    232 static void
    233 lwproc_makelwp(struct proc *p, struct lwp *l, bool doswitch, bool procmake)
    234 {
    235 
    236 	p->p_nlwps++;
    237 	l->l_refcnt = 1;
    238 	l->l_proc = p;
    239 
    240 	l->l_lid = p->p_nlwpid++;
    241 	LIST_INSERT_HEAD(&p->p_lwps, l, l_sibling);
    242 
    243 	l->l_fd = p->p_fd;
    244 	l->l_cpu = rump_cpu;
    245 	l->l_target_cpu = rump_cpu; /* Initial target CPU always the same */
    246 	l->l_stat = LSRUN;
    247 	l->l_mutex = &unruntime_lock;
    248 	TAILQ_INIT(&l->l_ld_locks);
    249 	mutex_exit(p->p_lock);
    250 
    251 	lwp_update_creds(l);
    252 	lwp_initspecific(l);
    253 
    254 	membar_enter();
    255 	rumpuser_curlwpop(RUMPUSER_LWP_CREATE, l);
    256 	if (doswitch) {
    257 		rump_lwproc_switch(l);
    258 	}
    259 
    260 	/* filedesc already has refcount 1 when process is created */
    261 	if (!procmake) {
    262 		fd_hold(l);
    263 	}
    264 
    265 	mutex_enter(proc_lock);
    266 	LIST_INSERT_HEAD(&alllwp, l, l_list);
    267 	mutex_exit(proc_lock);
    268 }
    269 
    270 struct lwp *
    271 rump__lwproc_alloclwp(struct proc *p)
    272 {
    273 	struct lwp *l;
    274 	bool newproc = false;
    275 
    276 	if (p == NULL) {
    277 		p = lwproc_newproc(&proc0, 0);
    278 		newproc = true;
    279 	}
    280 
    281 	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
    282 
    283 	mutex_enter(p->p_lock);
    284 	KASSERT((p->p_sflag & PS_RUMP_LWPEXIT) == 0);
    285 	lwproc_makelwp(p, l, false, newproc);
    286 
    287 	return l;
    288 }
    289 
    290 int
    291 rump_lwproc_newlwp(pid_t pid)
    292 {
    293 	struct proc *p;
    294 	struct lwp *l;
    295 
    296 	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
    297 	mutex_enter(proc_lock);
    298 	p = proc_find_raw(pid);
    299 	if (p == NULL) {
    300 		mutex_exit(proc_lock);
    301 		kmem_free(l, sizeof(*l));
    302 		return ESRCH;
    303 	}
    304 	mutex_enter(p->p_lock);
    305 	if (p->p_sflag & PS_RUMP_LWPEXIT) {
    306 		mutex_exit(proc_lock);
    307 		mutex_exit(p->p_lock);
    308 		kmem_free(l, sizeof(*l));
    309 		return EBUSY;
    310 	}
    311 	mutex_exit(proc_lock);
    312 	lwproc_makelwp(p, l, true, false);
    313 
    314 	return 0;
    315 }
    316 
    317 int
    318 rump_lwproc_rfork(int flags)
    319 {
    320 	struct proc *p;
    321 	struct lwp *l;
    322 
    323 	if (flags & ~(RUMP_RFFDG|RUMP_RFCFDG) ||
    324 	    (~flags & (RUMP_RFFDG|RUMP_RFCFDG)) == 0)
    325 		return EINVAL;
    326 
    327 	p = lwproc_newproc(curproc, flags);
    328 	l = kmem_zalloc(sizeof(*l), KM_SLEEP);
    329 	mutex_enter(p->p_lock);
    330 	KASSERT((p->p_sflag & PS_RUMP_LWPEXIT) == 0);
    331 	lwproc_makelwp(p, l, true, true);
    332 
    333 	return 0;
    334 }
    335 
    336 /*
    337  * Switch to a new process/thread.  Release previous one if
    338  * deemed to be exiting.  This is considered a slow path for
    339  * rump kernel entry.
    340  */
    341 void
    342 rump_lwproc_switch(struct lwp *newlwp)
    343 {
    344 	struct lwp *l = curlwp;
    345 
    346 	KASSERT(!(l->l_flag & LW_WEXIT) || newlwp);
    347 
    348 	if (__predict_false(newlwp && (newlwp->l_pflag & LP_RUNNING)))
    349 		panic("lwp %p (%d:%d) already running",
    350 		    newlwp, newlwp->l_proc->p_pid, newlwp->l_lid);
    351 
    352 	if (newlwp == NULL) {
    353 		l->l_pflag &= ~LP_RUNNING;
    354 		l->l_flag |= LW_RUMP_CLEAR;
    355 		return;
    356 	}
    357 
    358 	/* fd_free() must be called from curlwp context.  talk about ugh */
    359 	if (l->l_flag & LW_WEXIT) {
    360 		fd_free();
    361 	}
    362 
    363 	KERNEL_UNLOCK_ALL(NULL, &l->l_biglocks);
    364 	rumpuser_curlwpop(RUMPUSER_LWP_CLEAR, l);
    365 
    366 	newlwp->l_cpu = newlwp->l_target_cpu = l->l_cpu;
    367 	newlwp->l_mutex = l->l_mutex;
    368 	newlwp->l_pflag |= LP_RUNNING;
    369 
    370 	rumpuser_curlwpop(RUMPUSER_LWP_SET, newlwp);
    371 	curcpu()->ci_curlwp = newlwp;
    372 	KERNEL_LOCK(newlwp->l_biglocks, NULL);
    373 
    374 	/*
    375 	 * Check if the thread should get a signal.  This is
    376 	 * mostly to satisfy the "record" rump sigmodel.
    377 	 */
    378 	mutex_enter(newlwp->l_proc->p_lock);
    379 	if (sigispending(newlwp, 0)) {
    380 		newlwp->l_flag |= LW_PENDSIG;
    381 	}
    382 	mutex_exit(newlwp->l_proc->p_lock);
    383 
    384 	l->l_mutex = &unruntime_lock;
    385 	l->l_pflag &= ~LP_RUNNING;
    386 	l->l_flag &= ~LW_PENDSIG;
    387 	l->l_stat = LSRUN;
    388 
    389 	if (l->l_flag & LW_WEXIT) {
    390 		lwproc_freelwp(l);
    391 	}
    392 }
    393 
    394 /*
    395  * Mark the current thread to be released upon return from
    396  * kernel.
    397  */
    398 void
    399 rump_lwproc_releaselwp(void)
    400 {
    401 	struct lwp *l = curlwp;
    402 
    403 	if (l->l_refcnt == 0 || l->l_flag & LW_WEXIT)
    404 		panic("releasing non-pertinent lwp");
    405 
    406 	rump__lwproc_lwprele();
    407 	KASSERT(l->l_refcnt == 0 && (l->l_flag & LW_WEXIT));
    408 }
    409 
    410 /*
    411  * In-kernel routines used to add and remove references for the
    412  * current thread.  The main purpose is to make it possible for
    413  * implicit threads to persist over scheduling operations in
    414  * rump kernel drivers.  Note that we don't need p_lock in a
    415  * rump kernel, since we do refcounting only for curlwp.
    416  */
    417 void
    418 rump__lwproc_lwphold(void)
    419 {
    420 	struct lwp *l = curlwp;
    421 
    422 	l->l_refcnt++;
    423 	l->l_flag &= ~LW_WEXIT;
    424 }
    425 
    426 void
    427 rump__lwproc_lwprele(void)
    428 {
    429 	struct lwp *l = curlwp;
    430 
    431 	l->l_refcnt--;
    432 	if (l->l_refcnt == 0)
    433 		l->l_flag |= LW_WEXIT;
    434 }
    435 
    436 struct lwp *
    437 rump_lwproc_curlwp(void)
    438 {
    439 	struct lwp *l = curlwp;
    440 
    441 	if (l->l_flag & LW_WEXIT)
    442 		return NULL;
    443 	return l;
    444 }
    445 
    446 /* this interface is under construction (like the proverbial 90's web page) */
    447 int rump_i_know_what_i_am_doing_with_sysents = 0;
    448 void
    449 rump_lwproc_sysent_usenative()
    450 {
    451 
    452 	if (!rump_i_know_what_i_am_doing_with_sysents)
    453 		panic("don't use rump_lwproc_sysent_usenative()");
    454 	curproc->p_emul = &emul_netbsd;
    455 }
    456