Home | History | Annotate | Line # | Download | only in rumpkern
      1  1.361        ad /*	$NetBSD: rump.c,v 1.361 2023/10/05 19:41:07 ad Exp $	*/
      2    1.1     pooka 
      3    1.1     pooka /*
      4  1.233     pooka  * Copyright (c) 2007-2011 Antti Kantee.  All Rights Reserved.
      5    1.1     pooka  *
      6    1.1     pooka  * Redistribution and use in source and binary forms, with or without
      7    1.1     pooka  * modification, are permitted provided that the following conditions
      8    1.1     pooka  * are met:
      9    1.1     pooka  * 1. Redistributions of source code must retain the above copyright
     10    1.1     pooka  *    notice, this list of conditions and the following disclaimer.
     11    1.1     pooka  * 2. Redistributions in binary form must reproduce the above copyright
     12    1.1     pooka  *    notice, this list of conditions and the following disclaimer in the
     13    1.1     pooka  *    documentation and/or other materials provided with the distribution.
     14    1.1     pooka  *
     15    1.1     pooka  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
     16    1.1     pooka  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
     17    1.1     pooka  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
     18    1.1     pooka  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     19    1.1     pooka  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     20    1.1     pooka  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     21    1.1     pooka  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     22    1.1     pooka  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     23    1.1     pooka  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     24    1.1     pooka  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     25    1.1     pooka  * SUCH DAMAGE.
     26    1.1     pooka  */
     27    1.1     pooka 
     28   1.78     pooka #include <sys/cdefs.h>
     29  1.361        ad __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.361 2023/10/05 19:41:07 ad Exp $");
     30  1.165     pooka 
     31  1.165     pooka #include <sys/systm.h>
     32  1.165     pooka #define ELFSIZE ARCH_ELFSIZE
     33   1.78     pooka 
     34    1.1     pooka #include <sys/param.h>
     35   1.64     pooka #include <sys/atomic.h>
     36   1.75     pooka #include <sys/buf.h>
     37   1.66     pooka #include <sys/callout.h>
     38   1.79     pooka #include <sys/conf.h>
     39   1.27     pooka #include <sys/cpu.h>
     40  1.144     pooka #include <sys/device.h>
     41  1.354  riastrad #include <sys/device_impl.h>
     42  1.100     pooka #include <sys/evcnt.h>
     43   1.98     pooka #include <sys/event.h>
     44  1.137     pooka #include <sys/exec_elf.h>
     45    1.1     pooka #include <sys/filedesc.h>
     46   1.72     pooka #include <sys/iostat.h>
     47    1.1     pooka #include <sys/kauth.h>
     48  1.245  pgoyette #include <sys/kcpuset.h>
     49   1.80     pooka #include <sys/kernel.h>
     50   1.14     pooka #include <sys/kmem.h>
     51   1.81     pooka #include <sys/kprintf.h>
     52  1.175     pooka #include <sys/kthread.h>
     53   1.82     pooka #include <sys/ksyms.h>
     54   1.81     pooka #include <sys/msgbuf.h>
     55   1.49    simonb #include <sys/module.h>
     56  1.338  pgoyette #include <sys/module_hook.h>
     57  1.234     pooka #include <sys/namei.h>
     58   1.71     pooka #include <sys/once.h>
     59   1.65     pooka #include <sys/percpu.h>
     60  1.139     pooka #include <sys/pipe.h>
     61  1.162     pooka #include <sys/pool.h>
     62    1.1     pooka #include <sys/queue.h>
     63  1.121     pooka #include <sys/reboot.h>
     64    1.1     pooka #include <sys/resourcevar.h>
     65   1.27     pooka #include <sys/select.h>
     66   1.87     pooka #include <sys/sysctl.h>
     67   1.96     pooka #include <sys/syscall.h>
     68  1.169     pooka #include <sys/syscallvar.h>
     69  1.332   thorpej #include <sys/threadpool.h>
     70  1.160     pooka #include <sys/timetc.h>
     71   1.87     pooka #include <sys/tty.h>
     72   1.68     pooka #include <sys/uidinfo.h>
     73   1.89     pooka #include <sys/vmem.h>
     74  1.143     pooka #include <sys/xcall.h>
     75  1.235       tls #include <sys/cprng.h>
     76  1.317  riastrad #include <sys/rnd.h>
     77  1.279     pooka #include <sys/ktrace.h>
     78  1.337  riastrad #include <sys/pserialize.h>
     79  1.334     ozaki #include <sys/psref.h>
     80    1.1     pooka 
     81  1.327     pooka #include <rump-sys/kern.h>
     82  1.327     pooka #include <rump-sys/dev.h>
     83  1.327     pooka #include <rump-sys/net.h>
     84  1.327     pooka #include <rump-sys/vfs.h>
     85  1.327     pooka 
     86   1.48     pooka #include <rump/rumpuser.h>
     87   1.48     pooka 
     88  1.114     pooka #include <prop/proplib.h>
     89  1.114     pooka 
     90  1.174     pooka #include <uvm/uvm_extern.h>
     91  1.136     pooka #include <uvm/uvm_readahead.h>
     92  1.136     pooka 
     93  1.164     pooka char machine[] = MACHINE;
     94  1.322    justin char machine_arch[] = MACHINE_ARCH;
     95    1.1     pooka 
     96  1.193     pooka struct proc *initproc;
     97  1.101     pooka 
     98  1.144     pooka struct device rump_rootdev = {
     99  1.144     pooka 	.dv_class = DV_VIRTUAL
    100  1.144     pooka };
    101  1.144     pooka 
    102   1.54     pooka #ifdef RUMP_WITHOUT_THREADS
    103   1.54     pooka int rump_threads = 0;
    104   1.54     pooka #else
    105   1.54     pooka int rump_threads = 1;
    106   1.54     pooka #endif
    107   1.54     pooka 
    108  1.265     pooka static void rump_component_addlocal(void);
    109  1.255     pooka static struct lwp *bootlwp;
    110  1.255     pooka 
    111  1.329     joerg /* 16k should be enough for std rump needs */
    112  1.329     joerg static  char rump_msgbuf[16*1024] __aligned(256);
    113  1.156     pooka 
    114  1.218     pooka bool rump_ttycomponent = false;
    115  1.217     pooka 
    116  1.345     joerg extern pool_cache_t pnbuf_cache;
    117  1.333  christos 
    118   1.51     pooka static int rump_inited;
    119   1.27     pooka 
    120  1.299     pooka void (*rump_vfs_drainbufs)(int) = (void *)nullop;
    121  1.254     pooka int  (*rump_vfs_makeonedevnode)(dev_t, const char *,
    122  1.254     pooka 				devmajor_t, devminor_t) = (void *)nullop;
    123  1.254     pooka int  (*rump_vfs_makedevnodes)(dev_t, const char *, char,
    124  1.254     pooka 			      devmajor_t, devminor_t, int) = (void *)nullop;
    125  1.321     pooka int  (*rump_vfs_makesymlink)(const char *, const char *) = (void *)nullop;
    126  1.162     pooka 
    127  1.299     pooka rump_proc_vfs_init_fn rump_proc_vfs_init = (void *)nullop;
    128  1.299     pooka rump_proc_vfs_release_fn rump_proc_vfs_release = (void *)nullop;
    129   1.71     pooka 
    130  1.155     pooka static void add_linkedin_modules(const struct modinfo *const *, size_t);
    131  1.343  pgoyette static void add_static_evcnt(struct evcnt *);
    132  1.155     pooka 
    133  1.315     pooka static pid_t rspo_wrap_getpid(void) {
    134  1.315     pooka 	return rump_sysproxy_hyp_getpid();
    135  1.315     pooka }
    136  1.315     pooka static int rspo_wrap_syscall(int num, void *arg, long *retval) {
    137  1.315     pooka 	return rump_sysproxy_hyp_syscall(num, arg, retval);
    138  1.315     pooka }
    139  1.315     pooka static int rspo_wrap_rfork(void *priv, int flag, const char *comm) {
    140  1.315     pooka 	return rump_sysproxy_hyp_rfork(priv, flag, comm);
    141  1.315     pooka }
    142  1.315     pooka static void rspo_wrap_lwpexit(void) {
    143  1.315     pooka 	rump_sysproxy_hyp_lwpexit();
    144  1.315     pooka }
    145  1.315     pooka static void rspo_wrap_execnotify(const char *comm) {
    146  1.315     pooka 	rump_sysproxy_hyp_execnotify(comm);
    147  1.315     pooka }
    148  1.263     pooka static const struct rumpuser_hyperup hyp = {
    149  1.263     pooka 	.hyp_schedule		= rump_schedule,
    150  1.263     pooka 	.hyp_unschedule		= rump_unschedule,
    151  1.263     pooka 	.hyp_backend_unschedule	= rump_user_unschedule,
    152  1.263     pooka 	.hyp_backend_schedule	= rump_user_schedule,
    153  1.263     pooka 	.hyp_lwproc_switch	= rump_lwproc_switch,
    154  1.263     pooka 	.hyp_lwproc_release	= rump_lwproc_releaselwp,
    155  1.263     pooka 	.hyp_lwproc_newlwp	= rump_lwproc_newlwp,
    156  1.263     pooka 	.hyp_lwproc_curlwp	= rump_lwproc_curlwp,
    157  1.315     pooka 
    158  1.315     pooka 	.hyp_getpid		= rspo_wrap_getpid,
    159  1.315     pooka 	.hyp_syscall		= rspo_wrap_syscall,
    160  1.315     pooka 	.hyp_lwproc_rfork	= rspo_wrap_rfork,
    161  1.315     pooka 	.hyp_lwpexit		= rspo_wrap_lwpexit,
    162  1.315     pooka 	.hyp_execnotify		= rspo_wrap_execnotify,
    163  1.315     pooka };
    164  1.315     pooka struct rump_sysproxy_ops rump_sysproxy_ops = {
    165  1.315     pooka 	.rspo_copyin		= (void *)enxio,
    166  1.315     pooka 	.rspo_copyinstr 	= (void *)enxio,
    167  1.315     pooka 	.rspo_copyout	 	= (void *)enxio,
    168  1.315     pooka 	.rspo_copyoutstr 	= (void *)enxio,
    169  1.315     pooka 	.rspo_anonmmap 		= (void *)enxio,
    170  1.315     pooka 	.rspo_raise 		= (void *)enxio,
    171  1.315     pooka 	.rspo_fini 		= (void *)enxio,
    172  1.315     pooka 	.rspo_hyp_getpid 	= (void *)enxio,
    173  1.315     pooka 	.rspo_hyp_syscall 	= (void *)enxio,
    174  1.315     pooka 	.rspo_hyp_rfork 	= (void *)enxio,
    175  1.315     pooka 	.rspo_hyp_lwpexit 	= (void *)enxio,
    176  1.315     pooka 	.rspo_hyp_execnotify 	= (void *)enxio,
    177  1.197     pooka };
    178  1.197     pooka 
    179   1.61     pooka int
    180  1.207     pooka rump_daemonize_begin(void)
    181  1.207     pooka {
    182  1.207     pooka 
    183  1.207     pooka 	if (rump_inited)
    184  1.207     pooka 		return EALREADY;
    185  1.207     pooka 
    186  1.207     pooka 	return rumpuser_daemonize_begin();
    187  1.207     pooka }
    188  1.207     pooka 
    189  1.207     pooka int
    190  1.207     pooka rump_daemonize_done(int error)
    191  1.207     pooka {
    192  1.207     pooka 
    193  1.207     pooka 	return rumpuser_daemonize_done(error);
    194  1.207     pooka }
    195  1.207     pooka 
    196  1.314     pooka #ifdef RUMP_USE_CTOR
    197  1.314     pooka 
    198  1.314     pooka /* sysctl bootstrap handling */
    199  1.314     pooka struct sysctl_boot_chain sysctl_boot_chain \
    200  1.314     pooka     = LIST_HEAD_INITIALIZER(sysctl_boot_chain);
    201  1.314     pooka __link_set_add_text(sysctl_funcs,voidop); /* ensure linkset is non-empty */
    202  1.314     pooka 
    203  1.314     pooka #else /* RUMP_USE_CTOR */
    204  1.314     pooka 
    205  1.252     pooka RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT)
    206  1.252     pooka {
    207  1.265     pooka 	__link_set_decl(rump_components, struct rump_component);
    208  1.252     pooka 
    209  1.252     pooka 	/*
    210  1.253     pooka 	 * Trick compiler into generating references so that statically
    211  1.253     pooka 	 * linked rump kernels are generated with the link set symbols.
    212  1.252     pooka 	 */
    213  1.253     pooka 	asm("" :: "r"(__start_link_set_rump_components));
    214  1.253     pooka 	asm("" :: "r"(__stop_link_set_rump_components));
    215  1.252     pooka }
    216  1.314     pooka 
    217  1.314     pooka #endif /* RUMP_USE_CTOR */
    218  1.252     pooka 
    219  1.207     pooka int
    220  1.350  christos rump_init_callback(void (*cpuinit_callback) (void))
    221    1.1     pooka {
    222   1.36     pooka 	char buf[256];
    223  1.339   thorpej 	struct timespec bts;
    224  1.268     pooka 	int64_t sec;
    225  1.268     pooka 	long nsec;
    226  1.274     pooka 	struct lwp *l, *initlwp;
    227  1.166     pooka 	int i, numcpu;
    228    1.1     pooka 
    229  1.121     pooka 	/* not reentrant */
    230   1.27     pooka 	if (rump_inited)
    231   1.61     pooka 		return 0;
    232  1.121     pooka 	else if (rump_inited == -1)
    233  1.350  christos 		panic("%s: host process restart required", __func__);
    234  1.121     pooka 	else
    235  1.121     pooka 		rump_inited = 1;
    236   1.27     pooka 
    237  1.258     pooka 	/* initialize hypervisor */
    238  1.263     pooka 	if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) {
    239  1.258     pooka 		rumpuser_dprintf("rumpuser init failed\n");
    240  1.258     pooka 		return EINVAL;
    241  1.177     pooka 	}
    242  1.177     pooka 
    243  1.276     njoly 	/* init minimal lwp/cpu context */
    244  1.290     pooka 	rump_lwproc_init();
    245  1.276     njoly 	l = &lwp0;
    246  1.328     pooka 	l->l_cpu = l->l_target_cpu = &rump_bootcpu;
    247  1.290     pooka 	rump_lwproc_curlwp_set(l);
    248  1.276     njoly 
    249  1.258     pooka 	/* retrieve env vars which affect the early stage of bootstrap */
    250  1.264     pooka 	if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) {
    251  1.258     pooka 		rump_threads = *buf != '0';
    252  1.258     pooka 	}
    253  1.264     pooka 	if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) {
    254  1.149     pooka 		if (*buf != '0')
    255  1.149     pooka 			boothowto = AB_VERBOSE;
    256  1.149     pooka 	}
    257  1.264     pooka 
    258  1.264     pooka 	if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0)
    259  1.350  christos 		panic("%s: mandatory hypervisor configuration (NCPU) missing",
    260  1.350  christos 		    __func__);
    261  1.264     pooka 	numcpu = strtoll(buf, NULL, 10);
    262  1.264     pooka 	if (numcpu < 1) {
    263  1.350  christos 		panic("%s: rump kernels are not lightweight enough for %d CPUs",
    264  1.350  christos 		    __func__, numcpu);
    265  1.167     pooka 	}
    266  1.258     pooka 
    267  1.256     pooka 	rump_thread_init();
    268  1.204     pooka 	rump_cpus_bootstrap(&numcpu);
    269  1.165     pooka 
    270  1.266     pooka 	rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec);
    271  1.339   thorpej 	bts.tv_sec = sec;
    272  1.339   thorpej 	bts.tv_nsec = nsec;
    273  1.157     pooka 
    274  1.156     pooka 	initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
    275  1.156     pooka 	aprint_verbose("%s%s", copyright, version);
    276  1.148     pooka 
    277  1.187     pooka 	rump_intr_init(numcpu);
    278   1.36     pooka 
    279  1.271     njoly 	rump_tsleep_init();
    280  1.271     njoly 
    281  1.261     pooka 	rumpuser_mutex_init(&rump_giantlock, RUMPUSER_MTX_SPIN);
    282   1.82     pooka 	ksyms_init();
    283  1.174     pooka 	uvm_init();
    284  1.100     pooka 	evcnt_init();
    285   1.89     pooka 
    286  1.246  pgoyette 	kcpuset_sysinit();
    287   1.89     pooka 	once_init();
    288  1.181  pgoyette 	kernconfig_lock_init();
    289  1.114     pooka 	prop_kern_init();
    290   1.63     pooka 
    291   1.52     pooka 	kmem_init();
    292  1.102     pooka 
    293  1.136     pooka 	uvm_ra_init();
    294  1.185     pooka 	uao_init();
    295  1.136     pooka 
    296  1.143     pooka 	callout_startup();
    297  1.133     pooka 
    298   1.81     pooka 	kprintf_init();
    299  1.330     ozaki 	percpu_init();
    300  1.337  riastrad 	pserialize_init();
    301   1.52     pooka 
    302   1.59     pooka 	kauth_init();
    303   1.59     pooka 
    304  1.238       jym 	secmodel_init();
    305  1.283     pooka 	sysctl_init();
    306  1.314     pooka 	/*
    307  1.314     pooka 	 * The above call to sysctl_init() only initializes sysctl nodes
    308  1.314     pooka 	 * from link sets.  Initialize sysctls in case we used ctors.
    309  1.314     pooka 	 */
    310  1.314     pooka #ifdef RUMP_USE_CTOR
    311  1.314     pooka 	{
    312  1.314     pooka 		struct sysctl_setup_chain *ssc;
    313  1.314     pooka 
    314  1.326     pooka 		while ((ssc = LIST_FIRST(&sysctl_boot_chain)) != NULL) {
    315  1.326     pooka 			LIST_REMOVE(ssc, ssc_entries);
    316  1.314     pooka 			ssc->ssc_func(NULL);
    317  1.314     pooka 		}
    318  1.314     pooka 	}
    319  1.314     pooka #endif /* RUMP_USE_CTOR */
    320  1.238       jym 
    321  1.240     njoly 	rnd_init();
    322  1.282     pooka 	rump_hyperentropy_init();
    323  1.236     njoly 
    324  1.184     pooka 	procinit();
    325  1.184     pooka 	proc0_init();
    326  1.192     pooka 	uid_init();
    327  1.192     pooka 	chgproccnt(0, 1);
    328  1.184     pooka 
    329  1.178     pooka 	l->l_proc = &proc0;
    330  1.361        ad 	l->l_cred = kauth_cred_hold(l->l_proc->p_cred);
    331  1.176     pooka 
    332  1.159     pooka 	lwpinit_specificdata();
    333  1.176     pooka 	lwp_initspecific(&lwp0);
    334   1.36     pooka 
    335  1.334     ozaki 	/* Must be called after lwpinit_specificdata */
    336  1.334     ozaki 	psref_init();
    337  1.334     ozaki 
    338  1.332   thorpej 	threadpools_init();
    339  1.332   thorpej 
    340  1.320     pooka 	loginit();
    341  1.320     pooka 
    342  1.247     pooka 	rump_biglock_init();
    343  1.247     pooka 
    344  1.186     pooka 	rump_scheduler_init(numcpu);
    345  1.193     pooka 	/* revert temporary context and schedule a semireal context */
    346  1.290     pooka 	rump_lwproc_curlwp_clear(l);
    347  1.193     pooka 	initproc = &proc0; /* borrow proc0 before we get initproc started */
    348  1.124     pooka 	rump_schedule();
    349  1.255     pooka 	bootlwp = curlwp;
    350  1.124     pooka 
    351  1.160     pooka 	inittimecounter();
    352  1.160     pooka 	ntp_init();
    353  1.160     pooka 
    354  1.280     pooka #ifdef KTRACE
    355  1.278     pooka 	ktrinit();
    356  1.280     pooka #endif
    357  1.278     pooka 
    358  1.339   thorpej 	tc_setclock(&bts);
    359  1.150     pooka 
    360  1.293     pooka 	extern krwlock_t exec_lock;
    361  1.293     pooka 	rw_init(&exec_lock);
    362  1.293     pooka 
    363  1.143     pooka 	/* we are mostly go.  do per-cpu subsystem init */
    364  1.186     pooka 	for (i = 0; i < numcpu; i++) {
    365  1.143     pooka 		struct cpu_info *ci = cpu_lookup(i);
    366  1.143     pooka 
    367  1.186     pooka 		/* attach non-bootstrap CPUs */
    368  1.186     pooka 		if (i > 0) {
    369  1.186     pooka 			rump_cpu_attach(ci);
    370  1.186     pooka 			ncpu++;
    371  1.186     pooka 		}
    372  1.340  riastrad 		snprintf(ci->ci_cpuname, sizeof ci->ci_cpuname, "cpu%d", i);
    373  1.186     pooka 
    374  1.143     pooka 		callout_init_cpu(ci);
    375  1.143     pooka 		softint_init(ci);
    376  1.143     pooka 		xc_init_cpu(ci);
    377  1.143     pooka 		pool_cache_cpu_init(ci);
    378  1.143     pooka 		selsysinit(ci);
    379  1.150     pooka 		percpu_init_cpu(ci);
    380  1.182     pooka 
    381  1.216     pooka 		TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
    382  1.216     pooka 		__cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
    383  1.216     pooka 
    384  1.182     pooka 		aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
    385  1.143     pooka 	}
    386  1.324     pooka 	ncpuonline = ncpu;
    387   1.43     pooka 
    388  1.309  riastrad 	/* Once all CPUs are detected, initialize the per-CPU cprng_fast.  */
    389  1.347  riastrad 	cprng_init();
    390  1.309  riastrad 	cprng_fast_init();
    391  1.309  riastrad 
    392  1.357  riastrad 	mp_online = true;
    393  1.357  riastrad 
    394  1.350  christos 	if (cpuinit_callback)
    395  1.350  christos 		(*cpuinit_callback)();
    396  1.350  christos 
    397  1.357  riastrad 	/* CPUs are up.  allow kernel threads to run */
    398  1.357  riastrad 	rump_thread_allow(NULL);
    399  1.357  riastrad 
    400  1.282     pooka 	rnd_init_softint();
    401  1.282     pooka 
    402   1.98     pooka 	kqueue_init();
    403   1.72     pooka 	iostat_init();
    404   1.43     pooka 	fd_sys_init();
    405   1.44        ad 	module_init();
    406  1.338  pgoyette 	module_hook_init();
    407   1.79     pooka 	devsw_init();
    408  1.139     pooka 	pipe_init();
    409  1.162     pooka 	resource_init();
    410  1.224     pooka 	procinit_sysctl();
    411  1.292     pooka 	time_init();
    412  1.342        ad 	config_init();
    413    1.1     pooka 
    414  1.175     pooka 	/* start page baroness */
    415  1.175     pooka 	if (rump_threads) {
    416  1.175     pooka 		if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
    417  1.175     pooka 		    uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
    418  1.350  christos 			panic("%s: pagedaemon create failed", __func__);
    419  1.175     pooka 	} else
    420  1.175     pooka 		uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
    421  1.175     pooka 
    422  1.175     pooka 	/* process dso's */
    423  1.255     pooka 	rumpuser_dl_bootstrap(add_linkedin_modules,
    424  1.343  pgoyette 	    rump_kernelfsym_load, rump_component_load, add_static_evcnt);
    425  1.155     pooka 
    426  1.265     pooka 	rump_component_addlocal();
    427  1.179     pooka 	rump_component_init(RUMP_COMPONENT_KERN);
    428  1.179     pooka 
    429  1.249     pooka 	/* initialize factions, if present */
    430  1.249     pooka 	rump_component_init(RUMP__FACTION_VFS);
    431  1.249     pooka 	/* pnbuf_cache is used even without vfs */
    432  1.249     pooka 	if (rump_component_count(RUMP__FACTION_VFS) == 0) {
    433  1.249     pooka 		pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
    434  1.249     pooka 		    NULL, IPL_NONE, NULL, NULL, NULL);
    435  1.249     pooka 	}
    436  1.249     pooka 	rump_component_init(RUMP__FACTION_NET);
    437  1.249     pooka 	rump_component_init(RUMP__FACTION_DEV);
    438  1.249     pooka 	KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1
    439  1.249     pooka 	    && rump_component_count(RUMP__FACTION_NET) <= 1
    440  1.249     pooka 	    && rump_component_count(RUMP__FACTION_DEV) <= 1);
    441  1.179     pooka 
    442  1.179     pooka 	rump_component_init(RUMP_COMPONENT_KERN_VFS);
    443  1.179     pooka 
    444  1.218     pooka 	/*
    445  1.218     pooka 	 * if we initialized the tty component above, the tyttymtx is
    446  1.218     pooka 	 * now initialized.  otherwise, we need to initialize it.
    447  1.218     pooka 	 */
    448  1.218     pooka 	if (!rump_ttycomponent)
    449  1.218     pooka 		mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
    450  1.218     pooka 
    451  1.112     pooka 	cold = 0;
    452   1.31     pooka 
    453  1.115     pooka 	sysctl_finalize();
    454  1.115     pooka 
    455  1.155     pooka 	module_init_class(MODULE_CLASS_ANY);
    456  1.140     pooka 
    457  1.264     pooka 	if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME,
    458  1.264     pooka 	    hostname, MAXHOSTNAMELEN) != 0) {
    459  1.350  christos 		panic(
    460  1.350  christos 		    "%s: mandatory hypervisor configuration (HOSTNAME) missing",
    461  1.350  christos 		    __func__);
    462  1.264     pooka 	}
    463    1.1     pooka 	hostnamelen = strlen(hostname);
    464   1.24     pooka 
    465   1.24     pooka 	sigemptyset(&sigcantmask);
    466   1.27     pooka 
    467   1.89     pooka 	if (rump_threads)
    468   1.89     pooka 		vmem_rehash_start();
    469   1.89     pooka 
    470  1.193     pooka 	/*
    471  1.274     pooka 	 * Create init (proc 1), used to attach implicit threads in rump.
    472  1.193     pooka 	 * (note: must be done after vfsinit to get cwdi)
    473  1.193     pooka 	 */
    474  1.274     pooka 	initlwp = rump__lwproc_alloclwp(NULL);
    475  1.348        ad 	mutex_enter(&proc_lock);
    476  1.193     pooka 	initproc = proc_find_raw(1);
    477  1.348        ad 	mutex_exit(&proc_lock);
    478  1.193     pooka 	if (initproc == NULL)
    479  1.350  christos 		panic("%s: where in the world is initproc?", __func__);
    480  1.318     pooka 	strlcpy(initproc->p_comm, "rumplocal", sizeof(initproc->p_comm));
    481  1.193     pooka 
    482  1.304     pooka 	rump_component_init(RUMP_COMPONENT_POSTINIT);
    483  1.213     pooka 
    484  1.304     pooka 	/* load syscalls */
    485  1.304     pooka 	rump_component_init(RUMP_COMPONENT_SYSCALL);
    486  1.249     pooka 
    487  1.274     pooka 	/* component inits done */
    488  1.274     pooka 	bootlwp = NULL;
    489  1.274     pooka 
    490  1.274     pooka 	/* open 0/1/2 for init */
    491  1.275     pooka 	KASSERT(rump_lwproc_curlwp() == NULL);
    492  1.274     pooka 	rump_lwproc_switch(initlwp);
    493  1.274     pooka 	rump_consdev_init();
    494  1.275     pooka 	rump_lwproc_switch(NULL);
    495  1.274     pooka 
    496  1.193     pooka 	/* release cpu */
    497  1.124     pooka 	rump_unschedule();
    498  1.116     pooka 
    499    1.2     pooka 	return 0;
    500    1.2     pooka }
    501  1.350  christos 
    502  1.350  christos int
    503  1.350  christos rump_init(void)
    504  1.350  christos {
    505  1.350  christos 	return rump_init_callback(NULL);
    506  1.350  christos }
    507  1.350  christos 
    508  1.259     pooka /* historic compat */
    509  1.259     pooka __strong_alias(rump__init,rump_init);
    510    1.2     pooka 
    511  1.154     pooka static int compcounter[RUMP_COMPONENT_MAX];
    512  1.248     pooka static int compinited[RUMP_COMPONENT_MAX];
    513  1.154     pooka 
    514  1.255     pooka /*
    515  1.255     pooka  * Yea, this is O(n^2), but we're only looking at a handful of components.
    516  1.255     pooka  * Components are always initialized from the thread that called rump_init().
    517  1.255     pooka  */
    518  1.295     pooka static LIST_HEAD(, rump_component) rchead = LIST_HEAD_INITIALIZER(rchead);
    519  1.255     pooka 
    520  1.296     pooka #ifdef RUMP_USE_CTOR
    521  1.296     pooka struct modinfo_boot_chain modinfo_boot_chain \
    522  1.296     pooka     = LIST_HEAD_INITIALIZER(modinfo_boot_chain);
    523  1.296     pooka 
    524  1.296     pooka static void
    525  1.296     pooka rump_component_addlocal(void)
    526  1.296     pooka {
    527  1.296     pooka 	struct modinfo_chain *mc;
    528  1.296     pooka 
    529  1.296     pooka 	while ((mc = LIST_FIRST(&modinfo_boot_chain)) != NULL) {
    530  1.296     pooka 		LIST_REMOVE(mc, mc_entries);
    531  1.296     pooka 		module_builtin_add(&mc->mc_info, 1, false);
    532  1.296     pooka 	}
    533  1.296     pooka }
    534  1.296     pooka 
    535  1.296     pooka #else /* RUMP_USE_CTOR */
    536  1.296     pooka 
    537  1.265     pooka static void
    538  1.265     pooka rump_component_addlocal(void)
    539  1.265     pooka {
    540  1.265     pooka 	__link_set_decl(rump_components, struct rump_component);
    541  1.265     pooka 	struct rump_component *const *rc;
    542  1.265     pooka 
    543  1.265     pooka 	__link_set_foreach(rc, rump_components) {
    544  1.265     pooka 		rump_component_load(*rc);
    545  1.265     pooka 	}
    546  1.265     pooka }
    547  1.296     pooka #endif /* RUMP_USE_CTOR */
    548  1.265     pooka 
    549  1.296     pooka void
    550  1.295     pooka rump_component_load(const struct rump_component *rc_const)
    551  1.154     pooka {
    552  1.295     pooka 	struct rump_component *rc, *rc_iter;
    553  1.295     pooka 
    554  1.326     pooka 	/* time for rump component loading and unloading has passed */
    555  1.326     pooka 	if (!cold)
    556  1.326     pooka 		return;
    557  1.326     pooka 
    558  1.295     pooka 	/*
    559  1.295     pooka 	 * XXX: this is ok since the "const" was removed from the
    560  1.295     pooka 	 * definition of RUMP_COMPONENT().
    561  1.295     pooka 	 *
    562  1.295     pooka 	 * However, to preserve the hypercall interface, the const
    563  1.295     pooka 	 * remains here.  This can be fixed in the next hypercall revision.
    564  1.295     pooka 	 */
    565  1.295     pooka 	rc = __UNCONST(rc_const);
    566  1.154     pooka 
    567  1.307    justin 	KASSERT(!rump_inited || curlwp == bootlwp);
    568  1.248     pooka 
    569  1.295     pooka 	LIST_FOREACH(rc_iter, &rchead, rc_entries) {
    570  1.295     pooka 		if (rc_iter == rc)
    571  1.255     pooka 			return;
    572  1.154     pooka 	}
    573  1.255     pooka 
    574  1.295     pooka 	LIST_INSERT_HEAD(&rchead, rc, rc_entries);
    575  1.255     pooka 	KASSERT(rc->rc_type < RUMP_COMPONENT_MAX);
    576  1.255     pooka 	compcounter[rc->rc_type]++;
    577  1.154     pooka }
    578  1.154     pooka 
    579  1.326     pooka void
    580  1.326     pooka rump_component_unload(struct rump_component *rc)
    581  1.326     pooka {
    582  1.326     pooka 
    583  1.326     pooka 	/*
    584  1.326     pooka 	 * Checking for cold is enough because rump_init() both
    585  1.326     pooka 	 * flips it and handles component loading.
    586  1.326     pooka 	 */
    587  1.326     pooka 	if (!cold)
    588  1.326     pooka 		return;
    589  1.326     pooka 
    590  1.326     pooka 	LIST_REMOVE(rc, rc_entries);
    591  1.326     pooka }
    592  1.326     pooka 
    593  1.154     pooka int
    594  1.154     pooka rump_component_count(enum rump_component_type type)
    595  1.154     pooka {
    596  1.154     pooka 
    597  1.255     pooka 	KASSERT(curlwp == bootlwp);
    598  1.255     pooka 	KASSERT(type < RUMP_COMPONENT_MAX);
    599  1.154     pooka 	return compcounter[type];
    600  1.154     pooka }
    601  1.154     pooka 
    602  1.154     pooka void
    603  1.154     pooka rump_component_init(enum rump_component_type type)
    604  1.154     pooka {
    605  1.353       chs 	struct rump_component *rc, *rc_next, rc_marker;
    606  1.154     pooka 
    607  1.255     pooka 	KASSERT(curlwp == bootlwp);
    608  1.248     pooka 	KASSERT(!compinited[type]);
    609  1.353       chs 
    610  1.353       chs 	rc_marker.rc_type = RUMP_COMPONENT_MAX;
    611  1.353       chs 	rc_marker.rc_init = NULL;
    612  1.353       chs 	for (rc = LIST_FIRST(&rchead); rc != NULL; rc = rc_next) {
    613  1.296     pooka 		if (rc->rc_type == type) {
    614  1.353       chs 			LIST_INSERT_AFTER(rc, &rc_marker, rc_entries);
    615  1.255     pooka 			rc->rc_init();
    616  1.296     pooka 			LIST_REMOVE(rc, rc_entries);
    617  1.353       chs 			rc_next = LIST_NEXT(&rc_marker, rc_entries);
    618  1.353       chs 			LIST_REMOVE(&rc_marker, rc_entries);
    619  1.353       chs 		} else {
    620  1.353       chs 			rc_next = LIST_NEXT(rc, rc_entries);
    621  1.296     pooka 		}
    622  1.255     pooka 	}
    623  1.248     pooka 	compinited[type] = 1;
    624  1.154     pooka }
    625  1.154     pooka 
    626  1.155     pooka /*
    627  1.155     pooka  * Initialize a module which has already been loaded and linked
    628  1.155     pooka  * with dlopen(). This is fundamentally the same as a builtin module.
    629  1.296     pooka  *
    630  1.296     pooka  * XXX: this interface does not really work in the RUMP_USE_CTOR case,
    631  1.296     pooka  * but I'm not sure it's anything to cry about.  In feeling blue,
    632  1.296     pooka  * things could somehow be handled via modinfo_boot_chain.
    633  1.155     pooka  */
    634   1.76     pooka int
    635  1.155     pooka rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
    636   1.76     pooka {
    637   1.76     pooka 
    638  1.155     pooka 	return module_builtin_add(mip, nmodinfo, true);
    639  1.106     pooka }
    640  1.106     pooka 
    641  1.155     pooka /*
    642  1.155     pooka  * Finish module (flawless victory, fatality!).
    643  1.155     pooka  */
    644  1.106     pooka int
    645  1.155     pooka rump_module_fini(const struct modinfo *mi)
    646  1.106     pooka {
    647  1.120     pooka 
    648  1.155     pooka 	return module_builtin_remove(mi, true);
    649  1.155     pooka }
    650  1.155     pooka 
    651  1.155     pooka /*
    652  1.155     pooka  * Add loaded and linked module to the builtin list.  It will
    653  1.155     pooka  * later be initialized with module_init_class().
    654  1.155     pooka  */
    655  1.155     pooka 
    656  1.155     pooka static void
    657  1.155     pooka add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
    658  1.155     pooka {
    659  1.106     pooka 
    660  1.155     pooka 	module_builtin_add(mip, nmodinfo, false);
    661   1.76     pooka }
    662   1.76     pooka 
    663  1.343  pgoyette /*
    664  1.344  pgoyette  * Add an evcnt.
    665  1.343  pgoyette  */
    666  1.343  pgoyette static void
    667  1.343  pgoyette add_static_evcnt(struct evcnt *ev)
    668  1.343  pgoyette {
    669  1.343  pgoyette 
    670  1.343  pgoyette 	evcnt_attach_static(ev);
    671  1.343  pgoyette }
    672  1.343  pgoyette 
    673  1.137     pooka int
    674  1.137     pooka rump_kernelfsym_load(void *symtab, uint64_t symsize,
    675  1.137     pooka 	char *strtab, uint64_t strsize)
    676  1.137     pooka {
    677  1.137     pooka 	static int inited = 0;
    678  1.137     pooka 	Elf64_Ehdr ehdr;
    679  1.137     pooka 
    680  1.137     pooka 	if (inited)
    681  1.137     pooka 		return EBUSY;
    682  1.137     pooka 	inited = 1;
    683  1.137     pooka 
    684  1.137     pooka 	/*
    685  1.137     pooka 	 * Use 64bit header since it's bigger.  Shouldn't make a
    686  1.137     pooka 	 * difference, since we're passing in all zeroes anyway.
    687  1.137     pooka 	 */
    688  1.137     pooka 	memset(&ehdr, 0, sizeof(ehdr));
    689  1.137     pooka 	ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
    690  1.137     pooka 
    691  1.137     pooka 	return 0;
    692  1.137     pooka }
    693  1.137     pooka 
    694  1.191     pooka int
    695  1.130     pooka rump_boot_gethowto()
    696  1.130     pooka {
    697  1.130     pooka 
    698  1.130     pooka 	return boothowto;
    699  1.130     pooka }
    700  1.130     pooka 
    701  1.130     pooka void
    702  1.130     pooka rump_boot_sethowto(int howto)
    703  1.130     pooka {
    704  1.130     pooka 
    705  1.130     pooka 	boothowto = howto;
    706  1.130     pooka }
    707  1.130     pooka 
    708  1.109     pooka int
    709  1.124     pooka rump_getversion(void)
    710  1.109     pooka {
    711  1.109     pooka 
    712  1.109     pooka 	return __NetBSD_Version__;
    713  1.109     pooka }
    714  1.289     pooka /* compat */
    715  1.289     pooka __strong_alias(rump_pub_getversion,rump_getversion);
    716  1.170     pooka 
    717  1.170     pooka /*
    718  1.170     pooka  * Note: may be called unscheduled.  Not fully safe since no locking
    719  1.170     pooka  * of allevents (currently that's not even available).
    720  1.170     pooka  */
    721  1.170     pooka void
    722  1.170     pooka rump_printevcnts()
    723  1.170     pooka {
    724  1.170     pooka 	struct evcnt *ev;
    725  1.170     pooka 
    726  1.170     pooka 	TAILQ_FOREACH(ev, &allevents, ev_list)
    727  1.170     pooka 		rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
    728  1.170     pooka 		    ev->ev_group, ev->ev_name, ev->ev_count);
    729  1.170     pooka }
    730  1.184     pooka 
    731  1.184     pooka /*
    732  1.184     pooka  * If you use this interface ... well ... all bets are off.
    733  1.184     pooka  * The original purpose is for the p2k fs server library to be
    734  1.184     pooka  * able to use the same pid/lid for VOPs as the host kernel.
    735  1.184     pooka  */
    736  1.184     pooka void
    737  1.184     pooka rump_allbetsareoff_setid(pid_t pid, int lid)
    738  1.184     pooka {
    739  1.184     pooka 	struct lwp *l = curlwp;
    740  1.184     pooka 	struct proc *p = l->l_proc;
    741  1.184     pooka 
    742  1.184     pooka 	l->l_lid = lid;
    743  1.184     pooka 	p->p_pid = pid;
    744  1.184     pooka }
    745  1.251    martin 
    746  1.251    martin static void
    747  1.251    martin ipiemu(void *a1, void *a2)
    748  1.251    martin {
    749  1.251    martin 
    750  1.251    martin 	xc__highpri_intr(NULL);
    751  1.251    martin }
    752  1.251    martin 
    753  1.251    martin void
    754  1.251    martin rump_xc_highpri(struct cpu_info *ci)
    755  1.251    martin {
    756  1.251    martin 
    757  1.251    martin 	if (ci)
    758  1.251    martin 		xc_unicast(0, ipiemu, NULL, NULL, ci);
    759  1.251    martin 	else
    760  1.251    martin 		xc_broadcast(0, ipiemu, NULL, NULL);
    761  1.251    martin }
    762  1.272     pooka 
    763  1.272     pooka int
    764  1.272     pooka rump_syscall(int num, void *data, size_t dlen, register_t *retval)
    765  1.272     pooka {
    766  1.272     pooka 	struct proc *p;
    767  1.272     pooka 	struct emul *e;
    768  1.272     pooka 	struct sysent *callp;
    769  1.281     pooka 	const int *etrans = NULL;
    770  1.272     pooka 	int rv;
    771  1.272     pooka 
    772  1.272     pooka 	rump_schedule();
    773  1.272     pooka 	p = curproc;
    774  1.272     pooka 	e = p->p_emul;
    775  1.272     pooka #ifndef __HAVE_MINIMAL_EMUL
    776  1.349     kamil 	num &= e->e_nsysent - 1;
    777  1.349     kamil #else
    778  1.349     kamil 	num &= SYS_NSYSENT - 1;
    779  1.272     pooka #endif
    780  1.272     pooka 	callp = e->e_sysent + num;
    781  1.272     pooka 
    782  1.277     pooka 	rv = sy_invoke(callp, curlwp, data, retval, num);
    783  1.281     pooka 
    784  1.281     pooka 	/*
    785  1.281     pooka 	 * I hope that (!__HAVE_MINIMAL_EMUL || __HAVE_SYSCALL_INTERN) is
    786  1.281     pooka 	 * an invariant ...
    787  1.281     pooka 	 */
    788  1.281     pooka #if !defined(__HAVE_MINIMAL_EMUL)
    789  1.281     pooka 	etrans = e->e_errno;
    790  1.281     pooka #elif defined(__HAVE_SYSCALL_INTERN)
    791  1.281     pooka 	etrans = p->p_emuldata;
    792  1.281     pooka #endif
    793  1.281     pooka 
    794  1.281     pooka 	if (etrans) {
    795  1.281     pooka 		rv = etrans[rv];
    796  1.281     pooka 		/*
    797  1.281     pooka 		 * XXX: small hack since Linux etrans vectors on some
    798  1.281     pooka 		 * archs contain negative errnos, but rump_syscalls
    799  1.281     pooka 		 * uses the -1 + errno ABI.  Note that these
    800  1.281     pooka 		 * negative values are always the result of translation,
    801  1.281     pooka 		 * otherwise the above translation method would not
    802  1.281     pooka 		 * work very well.
    803  1.281     pooka 		 */
    804  1.281     pooka 		if (rv < 0)
    805  1.281     pooka 			rv = -rv;
    806  1.281     pooka 	}
    807  1.272     pooka 	rump_unschedule();
    808  1.272     pooka 
    809  1.272     pooka 	return rv;
    810  1.272     pooka }
    811  1.292     pooka 
    812  1.304     pooka void
    813  1.304     pooka rump_syscall_boot_establish(const struct rump_onesyscall *calls, size_t ncall)
    814  1.304     pooka {
    815  1.304     pooka 	struct sysent *callp;
    816  1.304     pooka 	size_t i;
    817  1.304     pooka 
    818  1.304     pooka 	for (i = 0; i < ncall; i++) {
    819  1.304     pooka 		callp = rump_sysent + calls[i].ros_num;
    820  1.304     pooka 		KASSERT(bootlwp != NULL
    821  1.335  christos 		    && callp->sy_call == (sy_call_t *)(void *)enosys);
    822  1.304     pooka 		callp->sy_call = calls[i].ros_handler;
    823  1.304     pooka 	}
    824  1.304     pooka }
    825  1.304     pooka 
    826  1.306     pooka struct rump_boot_etfs *ebstart;
    827  1.306     pooka void
    828  1.306     pooka rump_boot_etfs_register(struct rump_boot_etfs *eb)
    829  1.306     pooka {
    830  1.306     pooka 
    831  1.306     pooka 	/*
    832  1.306     pooka 	 * Could use atomics, but, since caller would need to synchronize
    833  1.306     pooka 	 * against calling rump_init() anyway, easier to just specify the
    834  1.306     pooka 	 * interface as "caller serializes".  This solve-by-specification
    835  1.306     pooka 	 * approach avoids the grey area of using atomics before rump_init()
    836  1.306     pooka 	 * runs.
    837  1.306     pooka 	 */
    838  1.306     pooka 	eb->_eb_next = ebstart;
    839  1.306     pooka 	eb->eb_status = -1;
    840  1.306     pooka 	ebstart = eb;
    841  1.306     pooka }
    842