1 /* $NetBSD: kern_fork.c,v 1.84 2001/02/26 21:14:20 lukem Exp $ */ 2 3 /* 4 * Copyright (c) 1982, 1986, 1989, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. All advertising materials mentioning features or use of this software 21 * must display the following acknowledgement: 22 * This product includes software developed by the University of 23 * California, Berkeley and its contributors. 24 * 4. Neither the name of the University nor the names of its contributors 25 * may be used to endorse or promote products derived from this software 26 * without specific prior written permission. 27 * 28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 38 * SUCH DAMAGE. 39 * 40 * @(#)kern_fork.c 8.8 (Berkeley) 2/14/95 41 */ 42 43 #include "opt_ktrace.h" 44 #include "opt_multiprocessor.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/map.h> 49 #include <sys/filedesc.h> 50 #include <sys/kernel.h> 51 #include <sys/malloc.h> 52 #include <sys/pool.h> 53 #include <sys/mount.h> 54 #include <sys/proc.h> 55 #include <sys/resourcevar.h> 56 #include <sys/vnode.h> 57 #include <sys/file.h> 58 #include <sys/acct.h> 59 #include <sys/ktrace.h> 60 #include <sys/vmmeter.h> 61 #include <sys/sched.h> 62 #include <sys/signalvar.h> 63 64 #include <sys/syscallargs.h> 65 66 #include <uvm/uvm_extern.h> 67 68 int nprocs = 1; /* process 0 */ 69 70 /*ARGSUSED*/ 71 int 72 sys_fork(struct proc *p, void *v, register_t *retval) 73 { 74 75 return (fork1(p, 0, SIGCHLD, NULL, 0, NULL, NULL, retval, NULL)); 76 } 77 78 /* 79 * vfork(2) system call compatible with 4.4BSD (i.e. BSD with Mach VM). 80 * Address space is not shared, but parent is blocked until child exit. 81 */ 82 /*ARGSUSED*/ 83 int 84 sys_vfork(struct proc *p, void *v, register_t *retval) 85 { 86 87 return (fork1(p, FORK_PPWAIT, SIGCHLD, NULL, 0, NULL, NULL, 88 retval, NULL)); 89 } 90 91 /* 92 * New vfork(2) system call for NetBSD, which implements original 3BSD vfork(2) 93 * semantics. Address space is shared, and parent is blocked until child exit. 94 */ 95 /*ARGSUSED*/ 96 int 97 sys___vfork14(struct proc *p, void *v, register_t *retval) 98 { 99 100 return (fork1(p, FORK_PPWAIT|FORK_SHAREVM, SIGCHLD, NULL, 0, 101 NULL, NULL, retval, NULL)); 102 } 103 104 int 105 fork1(struct proc *p1, int flags, int exitsig, void *stack, size_t stacksize, 106 void (*func)(void *), void *arg, register_t *retval, 107 struct proc **rnewprocp) 108 { 109 struct proc *p2, *tp; 110 uid_t uid; 111 int count, s; 112 vaddr_t uaddr; 113 static int nextpid, pidchecked; 114 115 /* 116 * Although process entries are dynamically created, we still keep 117 * a global limit on the maximum number we will create. Don't allow 118 * a nonprivileged user to use the last process; don't let root 119 * exceed the limit. The variable nprocs is the current number of 120 * processes, maxproc is the limit. 121 */ 122 uid = p1->p_cred->p_ruid; 123 if (__predict_false((nprocs >= maxproc - 1 && uid != 0) || 124 nprocs >= maxproc)) { 125 tablefull("proc", "increase kern.maxproc or NPROC"); 126 return (EAGAIN); 127 } 128 nprocs++; 129 130 /* 131 * Increment the count of procs running with this uid. Don't allow 132 * a nonprivileged user to exceed their current limit. 133 */ 134 count = chgproccnt(uid, 1); 135 if (__predict_false(uid != 0 && count > 136 p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 137 (void)chgproccnt(uid, -1); 138 nprocs--; 139 return (EAGAIN); 140 } 141 142 /* 143 * Allocate virtual address space for the U-area now, while it 144 * is still easy to abort the fork operation if we're out of 145 * kernel virtual address space. The actual U-area pages will 146 * be allocated and wired in vm_fork(). 147 */ 148 149 #ifndef USPACE_ALIGN 150 #define USPACE_ALIGN 0 151 #endif 152 153 uaddr = uvm_km_valloc_align(kernel_map, USPACE, USPACE_ALIGN); 154 if (__predict_false(uaddr == 0)) { 155 (void)chgproccnt(uid, -1); 156 nprocs--; 157 return (ENOMEM); 158 } 159 160 /* 161 * We are now committed to the fork. From here on, we may 162 * block on resources, but resource allocation may NOT fail. 163 */ 164 165 /* Allocate new proc. */ 166 p2 = pool_get(&proc_pool, PR_WAITOK); 167 168 /* 169 * Make a proc table entry for the new process. 170 * Start by zeroing the section of proc that is zero-initialized, 171 * then copy the section that is copied directly from the parent. 172 */ 173 memset(&p2->p_startzero, 0, 174 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 175 memcpy(&p2->p_startcopy, &p1->p_startcopy, 176 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 177 178 #if !defined(MULTIPROCESSOR) 179 /* 180 * In the single-processor case, all processes will always run 181 * on the same CPU. So, initialize the child's CPU to the parent's 182 * now. In the multiprocessor case, the child's CPU will be 183 * initialized in the low-level context switch code when the 184 * process runs. 185 */ 186 p2->p_cpu = p1->p_cpu; 187 #else 188 /* 189 * zero child's cpu pointer so we don't get trash. 190 */ 191 p2->p_cpu = NULL; 192 #endif /* ! MULTIPROCESSOR */ 193 194 /* 195 * Duplicate sub-structures as needed. 196 * Increase reference counts on shared objects. 197 * The p_stats and p_sigacts substructs are set in uvm_fork(). 198 */ 199 p2->p_flag = P_INMEM | (p1->p_flag & P_SUGID); 200 p2->p_emul = p1->p_emul; 201 202 if (p1->p_flag & P_PROFIL) 203 startprofclock(p2); 204 p2->p_cred = pool_get(&pcred_pool, PR_WAITOK); 205 memcpy(p2->p_cred, p1->p_cred, sizeof(*p2->p_cred)); 206 p2->p_cred->p_refcnt = 1; 207 crhold(p1->p_ucred); 208 209 /* bump references to the text vnode (for procfs) */ 210 p2->p_textvp = p1->p_textvp; 211 if (p2->p_textvp) 212 VREF(p2->p_textvp); 213 214 if (flags & FORK_SHAREFILES) 215 fdshare(p1, p2); 216 else 217 p2->p_fd = fdcopy(p1); 218 219 if (flags & FORK_SHARECWD) 220 cwdshare(p1, p2); 221 else 222 p2->p_cwdi = cwdinit(p1); 223 224 /* 225 * If p_limit is still copy-on-write, bump refcnt, 226 * otherwise get a copy that won't be modified. 227 * (If PL_SHAREMOD is clear, the structure is shared 228 * copy-on-write.) 229 */ 230 if (p1->p_limit->p_lflags & PL_SHAREMOD) 231 p2->p_limit = limcopy(p1->p_limit); 232 else { 233 p2->p_limit = p1->p_limit; 234 p2->p_limit->p_refcnt++; 235 } 236 237 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 238 p2->p_flag |= P_CONTROLT; 239 if (flags & FORK_PPWAIT) 240 p2->p_flag |= P_PPWAIT; 241 LIST_INSERT_AFTER(p1, p2, p_pglist); 242 p2->p_pptr = p1; 243 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling); 244 LIST_INIT(&p2->p_children); 245 246 callout_init(&p2->p_realit_ch); 247 callout_init(&p2->p_tsleep_ch); 248 249 #ifdef KTRACE 250 /* 251 * Copy traceflag and tracefile if enabled. 252 * If not inherited, these were zeroed above. 253 */ 254 if (p1->p_traceflag & KTRFAC_INHERIT) { 255 p2->p_traceflag = p1->p_traceflag; 256 if ((p2->p_tracep = p1->p_tracep) != NULL) 257 ktradref(p2); 258 } 259 #endif 260 261 #ifdef __HAVE_SYSCALL_INTERN 262 (*p2->p_emul->e_syscall_intern)(p2); 263 #endif 264 265 scheduler_fork_hook(p1, p2); 266 267 /* 268 * Create signal actions for the child process. 269 */ 270 sigactsinit(p2, p1, flags & FORK_SHARESIGS); 271 272 /* 273 * If emulation has process fork hook, call it now. 274 */ 275 if (p2->p_emul->e_proc_fork) 276 (*p2->p_emul->e_proc_fork)(p2, p1); 277 278 /* 279 * This begins the section where we must prevent the parent 280 * from being swapped. 281 */ 282 PHOLD(p1); 283 284 /* 285 * Finish creating the child process. It will return through a 286 * different path later. 287 */ 288 p2->p_addr = (struct user *)uaddr; 289 uvm_fork(p1, p2, (flags & FORK_SHAREVM) ? TRUE : FALSE, 290 stack, stacksize, 291 (func != NULL) ? func : child_return, 292 (arg != NULL) ? arg : p2); 293 294 /* 295 * BEGIN PID ALLOCATION. 296 */ 297 s = proclist_lock_write(); 298 299 /* 300 * Find an unused process ID. We remember a range of unused IDs 301 * ready to use (from nextpid+1 through pidchecked-1). 302 */ 303 nextpid++; 304 retry: 305 /* 306 * If the process ID prototype has wrapped around, 307 * restart somewhat above 0, as the low-numbered procs 308 * tend to include daemons that don't exit. 309 */ 310 if (nextpid >= PID_MAX) { 311 nextpid = 500; 312 pidchecked = 0; 313 } 314 if (nextpid >= pidchecked) { 315 const struct proclist_desc *pd; 316 317 pidchecked = PID_MAX; 318 /* 319 * Scan the process lists to check whether this pid 320 * is in use. Remember the lowest pid that's greater 321 * than nextpid, so we can avoid checking for a while. 322 */ 323 pd = proclists; 324 again: 325 LIST_FOREACH(tp, pd->pd_list, p_list) { 326 while (tp->p_pid == nextpid || 327 tp->p_pgrp->pg_id == nextpid || 328 tp->p_session->s_sid == nextpid) { 329 nextpid++; 330 if (nextpid >= pidchecked) 331 goto retry; 332 } 333 if (tp->p_pid > nextpid && pidchecked > tp->p_pid) 334 pidchecked = tp->p_pid; 335 336 if (tp->p_pgrp->pg_id > nextpid && 337 pidchecked > tp->p_pgrp->pg_id) 338 pidchecked = tp->p_pgrp->pg_id; 339 340 if (tp->p_session->s_sid > nextpid && 341 pidchecked > tp->p_session->s_sid) 342 pidchecked = tp->p_session->s_sid; 343 } 344 345 /* 346 * If there's another list, scan it. If we have checked 347 * them all, we've found one! 348 */ 349 pd++; 350 if (pd->pd_list != NULL) 351 goto again; 352 } 353 354 /* Record the pid we've allocated. */ 355 p2->p_pid = nextpid; 356 357 /* Record the signal to be delivered to the parent on exit. */ 358 p2->p_exitsig = exitsig; 359 360 /* 361 * Put the proc on allproc before unlocking PID allocation 362 * so that waiters won't grab it as soon as we unlock. 363 */ 364 365 p2->p_stat = SIDL; /* protect against others */ 366 p2->p_forw = p2->p_back = NULL; /* shouldn't be necessary */ 367 368 LIST_INSERT_HEAD(&allproc, p2, p_list); 369 370 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 371 372 /* 373 * END PID ALLOCATION. 374 */ 375 proclist_unlock_write(s); 376 377 /* 378 * Make child runnable, set start time, and add to run queue. 379 */ 380 SCHED_LOCK(s); 381 p2->p_stats->p_start = time; 382 p2->p_acflag = AFORK; 383 p2->p_stat = SRUN; 384 setrunqueue(p2); 385 SCHED_UNLOCK(s); 386 387 /* 388 * Now can be swapped. 389 */ 390 PRELE(p1); 391 392 /* 393 * Update stats now that we know the fork was successful. 394 */ 395 uvmexp.forks++; 396 if (flags & FORK_PPWAIT) 397 uvmexp.forks_ppwait++; 398 if (flags & FORK_SHAREVM) 399 uvmexp.forks_sharevm++; 400 401 /* 402 * Pass a pointer to the new process to the caller. 403 */ 404 if (rnewprocp != NULL) 405 *rnewprocp = p2; 406 407 #ifdef KTRACE 408 if (KTRPOINT(p2, KTR_EMUL)) 409 ktremul(p2); 410 #endif 411 412 /* 413 * Preserve synchronization semantics of vfork. If waiting for 414 * child to exec or exit, set P_PPWAIT on child, and sleep on our 415 * proc (in case of exit). 416 */ 417 if (flags & FORK_PPWAIT) 418 while (p2->p_flag & P_PPWAIT) 419 tsleep(p1, PWAIT, "ppwait", 0); 420 421 /* 422 * Return child pid to parent process, 423 * marking us as parent via retval[1]. 424 */ 425 if (retval != NULL) { 426 retval[0] = p2->p_pid; 427 retval[1] = 0; 428 } 429 430 return (0); 431 } 432 433 #if defined(MULTIPROCESSOR) 434 /* 435 * XXX This is a slight hack to get newly-formed processes to 436 * XXX acquire the kernel lock as soon as they run. 437 */ 438 void 439 proc_trampoline_mp(void) 440 { 441 struct proc *p; 442 443 p = curproc; 444 445 SCHED_ASSERT_UNLOCKED(); 446 KERNEL_PROC_LOCK(p); 447 } 448 #endif 449