1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * %sccs.include.redist.c% 11 * 12 * @(#)kern_fork.c 8.5 (Berkeley) 01/21/94 13 */ 14 15 #include <sys/param.h> 16 #include <sys/systm.h> 17 #include <sys/map.h> 18 #include <sys/filedesc.h> 19 #include <sys/kernel.h> 20 #include <sys/malloc.h> 21 #include <sys/proc.h> 22 #include <sys/resourcevar.h> 23 #include <sys/vnode.h> 24 #include <sys/file.h> 25 #include <sys/acct.h> 26 #include <sys/ktrace.h> 27 28 struct fork_args { 29 int dummy; 30 }; 31 /* ARGSUSED */ 32 fork(p, uap, retval) 33 struct proc *p; 34 struct fork_args *uap; 35 int retval[]; 36 { 37 38 return (fork1(p, 0, retval)); 39 } 40 41 /* ARGSUSED */ 42 vfork(p, uap, retval) 43 struct proc *p; 44 struct fork_args *uap; 45 int retval[]; 46 { 47 48 return (fork1(p, 1, retval)); 49 } 50 51 int nprocs = 1; /* process 0 */ 52 53 fork1(p1, isvfork, retval) 54 register struct proc *p1; 55 int isvfork, retval[]; 56 { 57 register struct proc *p2; 58 register uid_t uid; 59 struct proc *newproc; 60 struct proc **hash; 61 int count; 62 static int nextpid, pidchecked = 0; 63 64 /* 65 * Although process entries are dynamically created, we still keep 66 * a global limit on the maximum number we will create. Don't allow 67 * a nonprivileged user to bring the system within one of the global 68 * limit; don't let root exceed the limit. The variable nprocs is 69 * the current number of processes, maxproc is the limit. 70 */ 71 uid = p1->p_cred->p_ruid; 72 if (nprocs >= maxproc || uid == 0 && nprocs >= maxproc + 1) { 73 tablefull("proc"); 74 return (EAGAIN); 75 } 76 /* 77 * Increment the count of procs running with this uid. Don't allow 78 * a nonprivileged user to exceed their current limit. 79 */ 80 count = chgproccnt(uid, 1); 81 if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) { 82 (void)chgproccnt(uid, -1); 83 return (EAGAIN); 84 } 85 86 /* Allocate new proc. */ 87 MALLOC(newproc, struct proc *, sizeof(struct proc), M_PROC, M_WAITOK); 88 89 /* 90 * Find an unused process ID. We remember a range of unused IDs 91 * ready to use (from nextpid+1 through pidchecked-1). 92 */ 93 nextpid++; 94 retry: 95 /* 96 * If the process ID prototype has wrapped around, 97 * restart somewhat above 0, as the low-numbered procs 98 * tend to include daemons that don't exit. 99 */ 100 if (nextpid >= PID_MAX) { 101 nextpid = 100; 102 pidchecked = 0; 103 } 104 if (nextpid >= pidchecked) { 105 int doingzomb = 0; 106 107 pidchecked = PID_MAX; 108 /* 109 * Scan the active and zombie procs to check whether this pid 110 * is in use. Remember the lowest pid that's greater 111 * than nextpid, so we can avoid checking for a while. 112 */ 113 p2 = (struct proc *)allproc; 114 again: 115 for (; p2 != NULL; p2 = p2->p_next) { 116 while (p2->p_pid == nextpid || 117 p2->p_pgrp->pg_id == nextpid) { 118 nextpid++; 119 if (nextpid >= pidchecked) 120 goto retry; 121 } 122 if (p2->p_pid > nextpid && pidchecked > p2->p_pid) 123 pidchecked = p2->p_pid; 124 if (p2->p_pgrp->pg_id > nextpid && 125 pidchecked > p2->p_pgrp->pg_id) 126 pidchecked = p2->p_pgrp->pg_id; 127 } 128 if (!doingzomb) { 129 doingzomb = 1; 130 p2 = zombproc; 131 goto again; 132 } 133 } 134 135 136 /* 137 * Link onto allproc (this should probably be delayed). 138 * Heavy use of volatile here to prevent the compiler from 139 * rearranging code. Yes, it *is* terribly ugly, but at least 140 * it works. 141 */ 142 nprocs++; 143 p2 = newproc; 144 #define Vp2 ((volatile struct proc *)p2) 145 Vp2->p_stat = SIDL; /* protect against others */ 146 Vp2->p_pid = nextpid; 147 /* 148 * This is really: 149 * p2->p_next = allproc; 150 * allproc->p_prev = &p2->p_next; 151 * p2->p_prev = &allproc; 152 * allproc = p2; 153 * The assignment via allproc is legal since it is never NULL. 154 */ 155 *(volatile struct proc **)&Vp2->p_next = allproc; 156 *(volatile struct proc ***)&allproc->p_prev = 157 (volatile struct proc **)&Vp2->p_next; 158 *(volatile struct proc ***)&Vp2->p_prev = &allproc; 159 allproc = Vp2; 160 #undef Vp2 161 p2->p_forw = p2->p_back = NULL; /* shouldn't be necessary */ 162 163 /* Insert on the hash chain. */ 164 hash = &pidhash[PIDHASH(p2->p_pid)]; 165 p2->p_hash = *hash; 166 *hash = p2; 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 bzero(&p2->p_startzero, 174 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 175 bcopy(&p1->p_startcopy, &p2->p_startcopy, 176 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 177 178 /* 179 * Duplicate sub-structures as needed. 180 * Increase reference counts on shared objects. 181 * The p_stats and p_sigacts substructs are set in vm_fork. 182 */ 183 p2->p_flag = P_INMEM; 184 if (p1->p_flag & P_PROFIL) 185 startprofclock(p2); 186 MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred), 187 M_SUBPROC, M_WAITOK); 188 bcopy(p1->p_cred, p2->p_cred, sizeof(*p2->p_cred)); 189 p2->p_cred->p_refcnt = 1; 190 crhold(p1->p_ucred); 191 192 /* bump references to the text vnode (for procfs) */ 193 p2->p_textvp = p1->p_textvp; 194 if (p2->p_textvp) 195 VREF(p2->p_textvp); 196 197 p2->p_fd = fdcopy(p1); 198 /* 199 * If p_limit is still copy-on-write, bump refcnt, 200 * otherwise get a copy that won't be modified. 201 * (If PL_SHAREMOD is clear, the structure is shared 202 * copy-on-write.) 203 */ 204 if (p1->p_limit->p_lflags & PL_SHAREMOD) 205 p2->p_limit = limcopy(p1->p_limit); 206 else { 207 p2->p_limit = p1->p_limit; 208 p2->p_limit->p_refcnt++; 209 } 210 211 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 212 p2->p_flag |= P_CONTROLT; 213 if (isvfork) 214 p2->p_flag |= P_PPWAIT; 215 p2->p_pgrpnxt = p1->p_pgrpnxt; 216 p1->p_pgrpnxt = p2; 217 p2->p_pptr = p1; 218 p2->p_osptr = p1->p_cptr; 219 if (p1->p_cptr) 220 p1->p_cptr->p_ysptr = p2; 221 p1->p_cptr = p2; 222 #ifdef KTRACE 223 /* 224 * Copy traceflag and tracefile if enabled. 225 * If not inherited, these were zeroed above. 226 */ 227 if (p1->p_traceflag&KTRFAC_INHERIT) { 228 p2->p_traceflag = p1->p_traceflag; 229 if ((p2->p_tracep = p1->p_tracep) != NULL) 230 VREF(p2->p_tracep); 231 } 232 #endif 233 234 /* 235 * This begins the section where we must prevent the parent 236 * from being swapped. 237 */ 238 p1->p_flag |= P_NOSWAP; 239 /* 240 * Set return values for child before vm_fork, 241 * so they can be copied to child stack. 242 * We return parent pid, and mark as child in retval[1]. 243 * NOTE: the kernel stack may be at a different location in the child 244 * process, and thus addresses of automatic variables (including retval) 245 * may be invalid after vm_fork returns in the child process. 246 */ 247 retval[0] = p1->p_pid; 248 retval[1] = 1; 249 if (vm_fork(p1, p2, isvfork)) { 250 /* 251 * Child process. Set start time and get to work. 252 */ 253 (void) splclock(); 254 p2->p_stats->p_start = time; 255 (void) spl0(); 256 p2->p_acflag = AFORK; 257 return (0); 258 } 259 260 /* 261 * Make child runnable and add to run queue. 262 */ 263 (void) splhigh(); 264 p2->p_stat = SRUN; 265 setrunqueue(p2); 266 (void) spl0(); 267 268 /* 269 * Now can be swapped. 270 */ 271 p1->p_flag &= ~P_NOSWAP; 272 273 /* 274 * Preserve synchronization semantics of vfork. If waiting for 275 * child to exec or exit, set P_PPWAIT on child, and sleep on our 276 * proc (in case of exit). 277 */ 278 if (isvfork) 279 while (p2->p_flag & P_PPWAIT) 280 tsleep(p1, PWAIT, "ppwait", 0); 281 282 /* 283 * Return child pid to parent process, 284 * marking us as parent via retval[1]. 285 */ 286 retval[0] = p2->p_pid; 287 retval[1] = 0; 288 return (0); 289 } 290