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 8.7 (Berkeley) 08/22/94 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 use the last process; don't let root 68 * exceed the limit. The variable nprocs is the current number of 69 * processes, maxproc is the limit. 70 */ 71 uid = p1->p_cred->p_ruid; 72 if ((nprocs >= maxproc - 1 && uid != 0) || nprocs >= maxproc) { 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 = allproc.lh_first; 114 again: 115 for (; p2 != 0; p2 = p2->p_list.le_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.lh_first; 131 goto again; 132 } 133 } 134 135 136 nprocs++; 137 p2 = newproc; 138 p2->p_stat = SIDL; /* protect against others */ 139 p2->p_pid = nextpid; 140 LIST_INSERT_HEAD(&allproc, p2, p_list); 141 p2->p_forw = p2->p_back = NULL; /* shouldn't be necessary */ 142 LIST_INSERT_HEAD(PIDHASH(p2->p_pid), p2, p_hash); 143 144 /* 145 * Make a proc table entry for the new process. 146 * Start by zeroing the section of proc that is zero-initialized, 147 * then copy the section that is copied directly from the parent. 148 */ 149 bzero(&p2->p_startzero, 150 (unsigned) ((caddr_t)&p2->p_endzero - (caddr_t)&p2->p_startzero)); 151 bcopy(&p1->p_startcopy, &p2->p_startcopy, 152 (unsigned) ((caddr_t)&p2->p_endcopy - (caddr_t)&p2->p_startcopy)); 153 154 /* 155 * Duplicate sub-structures as needed. 156 * Increase reference counts on shared objects. 157 * The p_stats and p_sigacts substructs are set in vm_fork. 158 */ 159 p2->p_flag = P_INMEM; 160 if (p1->p_flag & P_PROFIL) 161 startprofclock(p2); 162 MALLOC(p2->p_cred, struct pcred *, sizeof(struct pcred), 163 M_SUBPROC, M_WAITOK); 164 bcopy(p1->p_cred, p2->p_cred, sizeof(*p2->p_cred)); 165 p2->p_cred->p_refcnt = 1; 166 crhold(p1->p_ucred); 167 168 /* bump references to the text vnode (for procfs) */ 169 p2->p_textvp = p1->p_textvp; 170 if (p2->p_textvp) 171 VREF(p2->p_textvp); 172 173 p2->p_fd = fdcopy(p1); 174 /* 175 * If p_limit is still copy-on-write, bump refcnt, 176 * otherwise get a copy that won't be modified. 177 * (If PL_SHAREMOD is clear, the structure is shared 178 * copy-on-write.) 179 */ 180 if (p1->p_limit->p_lflags & PL_SHAREMOD) 181 p2->p_limit = limcopy(p1->p_limit); 182 else { 183 p2->p_limit = p1->p_limit; 184 p2->p_limit->p_refcnt++; 185 } 186 187 if (p1->p_session->s_ttyvp != NULL && p1->p_flag & P_CONTROLT) 188 p2->p_flag |= P_CONTROLT; 189 if (isvfork) 190 p2->p_flag |= P_PPWAIT; 191 LIST_INSERT_AFTER(p1, p2, p_pglist); 192 p2->p_pptr = p1; 193 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling); 194 LIST_INIT(&p2->p_children); 195 196 #ifdef KTRACE 197 /* 198 * Copy traceflag and tracefile if enabled. 199 * If not inherited, these were zeroed above. 200 */ 201 if (p1->p_traceflag&KTRFAC_INHERIT) { 202 p2->p_traceflag = p1->p_traceflag; 203 if ((p2->p_tracep = p1->p_tracep) != NULL) 204 VREF(p2->p_tracep); 205 } 206 #endif 207 208 /* 209 * This begins the section where we must prevent the parent 210 * from being swapped. 211 */ 212 p1->p_flag |= P_NOSWAP; 213 /* 214 * Set return values for child before vm_fork, 215 * so they can be copied to child stack. 216 * We return parent pid, and mark as child in retval[1]. 217 * NOTE: the kernel stack may be at a different location in the child 218 * process, and thus addresses of automatic variables (including retval) 219 * may be invalid after vm_fork returns in the child process. 220 */ 221 retval[0] = p1->p_pid; 222 retval[1] = 1; 223 if (vm_fork(p1, p2, isvfork)) { 224 /* 225 * Child process. Set start time and get to work. 226 */ 227 (void) splclock(); 228 p2->p_stats->p_start = time; 229 (void) spl0(); 230 p2->p_acflag = AFORK; 231 return (0); 232 } 233 234 /* 235 * Make child runnable and add to run queue. 236 */ 237 (void) splhigh(); 238 p2->p_stat = SRUN; 239 setrunqueue(p2); 240 (void) spl0(); 241 242 /* 243 * Now can be swapped. 244 */ 245 p1->p_flag &= ~P_NOSWAP; 246 247 /* 248 * Preserve synchronization semantics of vfork. If waiting for 249 * child to exec or exit, set P_PPWAIT on child, and sleep on our 250 * proc (in case of exit). 251 */ 252 if (isvfork) 253 while (p2->p_flag & P_PPWAIT) 254 tsleep(p1, PWAIT, "ppwait", 0); 255 256 /* 257 * Return child pid to parent process, 258 * marking us as parent via retval[1]. 259 */ 260 retval[0] = p2->p_pid; 261 retval[1] = 0; 262 return (0); 263 } 264