1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)kern_fork.c 8.1 (Berkeley) 06/10/93 8 */ 9 10 #include <sys/param.h> 11 #include <sys/systm.h> 12 #include <sys/map.h> 13 #include <sys/filedesc.h> 14 #include <sys/kernel.h> 15 #include <sys/malloc.h> 16 #include <sys/proc.h> 17 #include <sys/resourcevar.h> 18 #include <sys/vnode.h> 19 #include <sys/file.h> 20 #include <sys/acct.h> 21 #include <sys/ktrace.h> 22 23 struct fork_args { 24 int dummy; 25 }; 26 /* ARGSUSED */ 27 fork(p, uap, retval) 28 struct proc *p; 29 struct fork_args *uap; 30 int retval[]; 31 { 32 33 return (fork1(p, 0, retval)); 34 } 35 36 /* ARGSUSED */ 37 vfork(p, uap, retval) 38 struct proc *p; 39 struct fork_args *uap; 40 int retval[]; 41 { 42 43 return (fork1(p, 1, retval)); 44 } 45 46 int nprocs = 1; /* process 0 */ 47 48 fork1(p1, isvfork, retval) 49 register struct proc *p1; 50 int isvfork, retval[]; 51 { 52 register struct proc *p2; 53 register uid_t uid; 54 struct proc *newproc; 55 struct proc **hash; 56 int count; 57 static int nextpid, pidchecked = 0; 58 59 /* 60 * Although process entries are dynamically created, we still keep 61 * a global limit on the maximum number we will create. Don't allow 62 * a nonprivileged user to bring the system within one of the global 63 * limit; don't let root exceed the limit. The variable nprocs is 64 * the current number of processes, maxproc is the limit. 65 */ 66 uid = p1->p_cred->p_ruid; 67 if (nprocs >= maxproc || uid == 0 && nprocs >= maxproc + 1) { 68 tablefull("proc"); 69 return (EAGAIN); 70 } 71 /* 72 * Increment the count of procs running with this uid. Don't allow 73 * a nonprivileged user to exceed their current limit. 74 */ 75 count = chgproccnt(uid, 1); 76 if (uid != 0 && count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur) { 77 (void)chgproccnt(uid, -1); 78 return (EAGAIN); 79 } 80 81 /* Allocate new proc. */ 82 MALLOC(newproc, struct proc *, sizeof(struct proc), M_PROC, M_WAITOK); 83 84 /* 85 * Find an unused process ID. We remember a range of unused IDs 86 * ready to use (from nextpid+1 through pidchecked-1). 87 */ 88 nextpid++; 89 retry: 90 /* 91 * If the process ID prototype has wrapped around, 92 * restart somewhat above 0, as the low-numbered procs 93 * tend to include daemons that don't exit. 94 */ 95 if (nextpid >= PID_MAX) { 96 nextpid = 100; 97 pidchecked = 0; 98 } 99 if (nextpid >= pidchecked) { 100 int doingzomb = 0; 101 102 pidchecked = PID_MAX; 103 /* 104 * Scan the active and zombie procs to check whether this pid 105 * is in use. Remember the lowest pid that's greater 106 * than nextpid, so we can avoid checking for a while. 107 */ 108 p2 = (struct proc *)allproc; 109 again: 110 for (; p2 != NULL; p2 = p2->p_nxt) { 111 while (p2->p_pid == nextpid || 112 p2->p_pgrp->pg_id == nextpid) { 113 nextpid++; 114 if (nextpid >= pidchecked) 115 goto retry; 116 } 117 if (p2->p_pid > nextpid && pidchecked > p2->p_pid) 118 pidchecked = p2->p_pid; 119 if (p2->p_pgrp->pg_id > nextpid && 120 pidchecked > p2->p_pgrp->pg_id) 121 pidchecked = p2->p_pgrp->pg_id; 122 } 123 if (!doingzomb) { 124 doingzomb = 1; 125 p2 = zombproc; 126 goto again; 127 } 128 } 129 130 131 /* 132 * Link onto allproc (this should probably be delayed). 133 * Heavy use of volatile here to prevent the compiler from 134 * rearranging code. Yes, it *is* terribly ugly, but at least 135 * it works. 136 */ 137 nprocs++; 138 p2 = newproc; 139 #define Vp2 ((volatile struct proc *)p2) 140 Vp2->p_stat = SIDL; /* protect against others */ 141 Vp2->p_pid = nextpid; 142 /* 143 * This is really: 144 * p2->p_nxt = allproc; 145 * allproc->p_prev = &p2->p_nxt; 146 * p2->p_prev = &allproc; 147 * allproc = p2; 148 * The assignment via allproc is legal since it is never NULL. 149 */ 150 *(volatile struct proc **)&Vp2->p_nxt = allproc; 151 *(volatile struct proc ***)&allproc->p_prev = 152 (volatile struct proc **)&Vp2->p_nxt; 153 *(volatile struct proc ***)&Vp2->p_prev = &allproc; 154 allproc = Vp2; 155 #undef Vp2 156 p2->p_link = NULL; /* shouldn't be necessary */ 157 p2->p_rlink = NULL; /* shouldn't be necessary */ 158 159 /* Insert on the hash chain. */ 160 hash = &pidhash[PIDHASH(p2->p_pid)]; 161 p2->p_hash = *hash; 162 *hash = p2; 163 164 /* 165 * Make a proc table entry for the new process. 166 * Start by zeroing the section of proc that is zero-initialized, 167 * then copy the section that is copied directly from the parent. 168 */ 169 bzero(&p2->p_startzero, 170 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 171 bcopy(&p1->p_startcopy, &p2->p_startcopy, 172 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 173 174 /* 175 * Duplicate sub-structures as needed. 176 * Increase reference counts on shared objects. 177 * The p_stats and p_sigacts substructs are set in vm_fork. 178 */ 179 p2->p_flag = SLOAD; 180 if (p1->p_flag & SPROFIL) 181 startprofclock(p2); 182 MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred), 183 M_SUBPROC, M_WAITOK); 184 bcopy(p1->p_cred, p2->p_cred, sizeof(*p2->p_cred)); 185 p2->p_cred->p_refcnt = 1; 186 crhold(p1->p_ucred); 187 188 p2->p_fd = fdcopy(p1); 189 /* 190 * If p_limit is still copy-on-write, bump refcnt, 191 * otherwise get a copy that won't be modified. 192 * (If PL_SHAREMOD is clear, the structure is shared 193 * copy-on-write.) 194 */ 195 if (p1->p_limit->p_lflags & PL_SHAREMOD) 196 p2->p_limit = limcopy(p1->p_limit); 197 else { 198 p2->p_limit = p1->p_limit; 199 p2->p_limit->p_refcnt++; 200 } 201 202 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & SCTTY) 203 p2->p_flag |= SCTTY; 204 if (isvfork) 205 p2->p_flag |= SPPWAIT; 206 p2->p_pgrpnxt = p1->p_pgrpnxt; 207 p1->p_pgrpnxt = p2; 208 p2->p_pptr = p1; 209 p2->p_osptr = p1->p_cptr; 210 if (p1->p_cptr) 211 p1->p_cptr->p_ysptr = p2; 212 p1->p_cptr = p2; 213 #ifdef KTRACE 214 /* 215 * Copy traceflag and tracefile if enabled. 216 * If not inherited, these were zeroed above. 217 */ 218 if (p1->p_traceflag&KTRFAC_INHERIT) { 219 p2->p_traceflag = p1->p_traceflag; 220 if ((p2->p_tracep = p1->p_tracep) != NULL) 221 VREF(p2->p_tracep); 222 } 223 #endif 224 225 /* 226 * This begins the section where we must prevent the parent 227 * from being swapped. 228 */ 229 p1->p_flag |= SKEEP; 230 /* 231 * Set return values for child before vm_fork, 232 * so they can be copied to child stack. 233 * We return parent pid, and mark as child in retval[1]. 234 * NOTE: the kernel stack may be at a different location in the child 235 * process, and thus addresses of automatic variables (including retval) 236 * may be invalid after vm_fork returns in the child process. 237 */ 238 retval[0] = p1->p_pid; 239 retval[1] = 1; 240 if (vm_fork(p1, p2, isvfork)) { 241 /* 242 * Child process. Set start time and get to work. 243 */ 244 (void) splclock(); 245 p2->p_stats->p_start = time; 246 (void) spl0(); 247 p2->p_acflag = AFORK; 248 return (0); 249 } 250 251 /* 252 * Make child runnable and add to run queue. 253 */ 254 (void) splhigh(); 255 p2->p_stat = SRUN; 256 setrq(p2); 257 (void) spl0(); 258 259 /* 260 * Now can be swapped. 261 */ 262 p1->p_flag &= ~SKEEP; 263 264 /* 265 * Preserve synchronization semantics of vfork. 266 * If waiting for child to exec or exit, set SPPWAIT 267 * on child, and sleep on our proc (in case of exit). 268 */ 269 if (isvfork) 270 while (p2->p_flag & SPPWAIT) 271 tsleep((caddr_t)p1, PWAIT, "ppwait", 0); 272 273 /* 274 * Return child pid to parent process, 275 * marking us as parent via retval[1]. 276 */ 277 retval[0] = p2->p_pid; 278 retval[1] = 0; 279 return (0); 280 } 281