2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/sysctl.h>
34 #include <sys/malloc.h>
36 #include <sys/vnode.h>
38 #include <sys/filedesc.h>
40 #include <sys/dsched.h>
41 #include <sys/signalvar.h>
42 #include <sys/spinlock.h>
43 #include <sys/random.h>
44 #include <sys/vnode.h>
48 #include <vm/vm_map.h>
50 #include <machine/smp.h>
52 #include <sys/refcount.h>
53 #include <sys/spinlock2.h>
54 #include <sys/mplock2.h>
57 * Hash table size must be a power of two and is not currently dynamically
58 * sized. There is a trade-off between the linear scans which must iterate
59 * all HSIZE elements and the number of elements which might accumulate
60 * within each hash chain.
62 #define ALLPROC_HSIZE 256
63 #define ALLPROC_HMASK (ALLPROC_HSIZE - 1)
64 #define ALLPROC_HASH(pid) (pid & ALLPROC_HMASK)
65 #define PGRP_HASH(pid) (pid & ALLPROC_HMASK)
66 #define SESS_HASH(pid) (pid & ALLPROC_HMASK)
69 * pid_doms[] management, used to control how quickly a PID can be recycled.
70 * Must be a multiple of ALLPROC_HSIZE for the proc_makepid() inner loops.
72 * WARNING! PIDDOM_DELAY should not be defined > 20 or so unless you change
73 * the array from int8_t's to int16_t's.
75 #define PIDDOM_COUNT 10 /* 10 pids per domain - reduce array size */
76 #define PIDDOM_DELAY 10 /* min 10 seconds after exit before reuse */
77 #define PIDSEL_DOMAINS (PID_MAX / PIDDOM_COUNT / ALLPROC_HSIZE * ALLPROC_HSIZE)
80 int allproc_hsize = ALLPROC_HSIZE;
82 LIST_HEAD(pidhashhead, proc);
84 static MALLOC_DEFINE(M_PGRP, "pgrp", "process group header");
85 MALLOC_DEFINE(M_SESSION, "session", "session header");
86 MALLOC_DEFINE(M_PROC, "proc", "Proc structures");
87 MALLOC_DEFINE(M_LWP, "lwp", "lwp structures");
88 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures");
90 int ps_showallprocs = 1;
91 static int ps_showallthreads = 1;
92 SYSCTL_INT(_security, OID_AUTO, ps_showallprocs, CTLFLAG_RW,
94 "Unprivileged processes can see processes with different UID/GID");
95 SYSCTL_INT(_security, OID_AUTO, ps_showallthreads, CTLFLAG_RW,
96 &ps_showallthreads, 0,
97 "Unprivileged processes can see kernel threads");
98 static u_int pid_domain_skips;
99 SYSCTL_UINT(_kern, OID_AUTO, pid_domain_skips, CTLFLAG_RW,
100 &pid_domain_skips, 0,
101 "Number of pid_doms[] skipped");
102 static u_int pid_inner_skips;
103 SYSCTL_UINT(_kern, OID_AUTO, pid_inner_skips, CTLFLAG_RW,
105 "Number of pid_doms[] skipped");
107 static void orphanpg(struct pgrp *pg);
108 static void proc_makepid(struct proc *p, int random_offset);
111 * Other process lists
113 static struct lwkt_token proc_tokens[ALLPROC_HSIZE];
114 static struct proclist allprocs[ALLPROC_HSIZE]; /* locked by proc_tokens */
115 static struct pgrplist allpgrps[ALLPROC_HSIZE]; /* locked by proc_tokens */
116 static struct sesslist allsessn[ALLPROC_HSIZE]; /* locked by proc_tokens */
119 * We try our best to avoid recycling a PID too quickly. We do this by
120 * storing (uint8_t)time_second in the related pid domain on-reap and then
121 * using that to skip-over the domain on-allocate.
123 * This array has to be fairly large to support a high fork/exec rate.
124 * We want ~100,000 entries or so to support a 10-second reuse latency
125 * at 10,000 execs/second, worst case. Best-case multiply by PIDDOM_COUNT
126 * (approximately 100,000 execs/second).
128 static uint8_t pid_doms[PIDSEL_DOMAINS]; /* ~100,000 entries */
131 * Random component to nextpid generation. We mix in a random factor to make
132 * it a little harder to predict. We sanity check the modulus value to avoid
133 * doing it in critical paths. Don't let it be too small or we pointlessly
134 * waste randomness entropy, and don't let it be impossibly large. Using a
135 * modulus that is too big causes a LOT more process table scans and slows
136 * down fork processing as the pidchecked caching is defeated.
138 static int randompid = 0;
144 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
149 error = sysctl_handle_int(oidp, &pid, 0, req);
150 if (error || !req->newptr)
152 if (pid < 0 || pid > PID_MAX - 100) /* out of range */
154 else if (pid < 2) /* NOP */
156 else if (pid < 100) /* Make it reasonable */
162 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
163 0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
166 * Initialize global process hashing structures.
168 * These functions are ONLY called from the low level boot code and do
169 * not lock their operations.
177 * Avoid unnecessary stalls due to pid_doms[] values all being
178 * the same. Make sure that the allocation of pid 1 and pid 2
181 for (i = 0; i < PIDSEL_DOMAINS; ++i)
182 pid_doms[i] = (int8_t)i - (int8_t)(PIDDOM_DELAY + 1);
187 for (i = 0; i < ALLPROC_HSIZE; ++i) {
188 LIST_INIT(&allprocs[i]);
189 LIST_INIT(&allsessn[i]);
190 LIST_INIT(&allpgrps[i]);
191 lwkt_token_init(&proc_tokens[i], "allproc");
197 procinsertinit(struct proc *p)
199 LIST_INSERT_HEAD(&allprocs[ALLPROC_HASH(p->p_pid)], p, p_list);
203 pgrpinsertinit(struct pgrp *pg)
205 LIST_INSERT_HEAD(&allpgrps[ALLPROC_HASH(pg->pg_id)], pg, pg_list);
209 sessinsertinit(struct session *sess)
211 LIST_INSERT_HEAD(&allsessn[ALLPROC_HASH(sess->s_sid)], sess, s_list);
215 * Process hold/release support functions. Called via the PHOLD(),
216 * PRELE(), and PSTALL() macros.
218 * p->p_lock is a simple hold count with a waiting interlock. No wakeup()
219 * is issued unless someone is actually waiting for the process.
221 * Most holds are short-term, allowing a process scan or other similar
222 * operation to access a proc structure without it getting ripped out from
223 * under us. procfs and process-list sysctl ops also use the hold function
224 * interlocked with various p_flags to keep the vmspace intact when reading
225 * or writing a user process's address space.
227 * There are two situations where a hold count can be longer. Exiting lwps
228 * hold the process until the lwp is reaped, and the parent will hold the
229 * child during vfork()/exec() sequences while the child is marked P_PPWAIT.
231 * The kernel waits for the hold count to drop to 0 (or 1 in some cases) at
232 * various critical points in the fork/exec and exit paths before proceeding.
234 #define PLOCK_ZOMB 0x20000000
235 #define PLOCK_WAITING 0x40000000
236 #define PLOCK_MASK 0x1FFFFFFF
239 pstall(struct proc *p, const char *wmesg, int count)
247 if ((o & PLOCK_MASK) <= count)
249 n = o | PLOCK_WAITING;
250 tsleep_interlock(&p->p_lock, 0);
253 * If someone is trying to single-step the process during
254 * an exec or an exit they can deadlock us because procfs
255 * sleeps with the process held.
258 if (p->p_flags & P_INEXEC) {
260 } else if (p->p_flags & P_POSTEXIT) {
261 spin_lock(&p->p_spin);
264 spin_unlock(&p->p_spin);
269 if (atomic_cmpset_int(&p->p_lock, o, n)) {
270 tsleep(&p->p_lock, PINTERLOCKED, wmesg, 0);
276 phold(struct proc *p)
278 atomic_add_int(&p->p_lock, 1);
282 * WARNING! On last release (p) can become instantly invalid due to
286 prele(struct proc *p)
294 if (atomic_cmpset_int(&p->p_lock, 1, 0))
302 KKASSERT((o & PLOCK_MASK) > 0);
304 n = (o - 1) & ~PLOCK_WAITING;
305 if (atomic_cmpset_int(&p->p_lock, o, n)) {
306 if (o & PLOCK_WAITING)
314 * Hold and flag serialized for zombie reaping purposes.
316 * This function will fail if it has to block, returning non-zero with
317 * neither the flag set or the hold count bumped. Note that we must block
318 * without holding a ref, meaning that the caller must ensure that (p)
319 * remains valid through some other interlock (typically on its parent
320 * process's p_token).
322 * Zero is returned on success. The hold count will be incremented and
323 * the serialization flag acquired. Note that serialization is only against
324 * other pholdzomb() calls, not against phold() calls.
327 pholdzomb(struct proc *p)
335 if (atomic_cmpset_int(&p->p_lock, 0, PLOCK_ZOMB | 1))
344 if ((o & PLOCK_ZOMB) == 0) {
345 n = (o + 1) | PLOCK_ZOMB;
346 if (atomic_cmpset_int(&p->p_lock, o, n))
349 KKASSERT((o & PLOCK_MASK) > 0);
350 n = o | PLOCK_WAITING;
351 tsleep_interlock(&p->p_lock, 0);
352 if (atomic_cmpset_int(&p->p_lock, o, n)) {
353 tsleep(&p->p_lock, PINTERLOCKED, "phldz", 0);
354 /* (p) can be ripped out at this point */
362 * Release PLOCK_ZOMB and the hold count, waking up any waiters.
364 * WARNING! On last release (p) can become instantly invalid due to
368 prelezomb(struct proc *p)
376 if (atomic_cmpset_int(&p->p_lock, PLOCK_ZOMB | 1, 0))
382 KKASSERT(p->p_lock & PLOCK_ZOMB);
385 KKASSERT((o & PLOCK_MASK) > 0);
387 n = (o - 1) & ~(PLOCK_ZOMB | PLOCK_WAITING);
388 if (atomic_cmpset_int(&p->p_lock, o, n)) {
389 if (o & PLOCK_WAITING)
397 * Is p an inferior of the current process?
402 inferior(struct proc *p)
407 lwkt_gettoken_shared(&p->p_token);
408 while (p != curproc) {
410 lwkt_reltoken(&p->p_token);
415 lwkt_reltoken(&p->p_token);
417 lwkt_gettoken_shared(&p2->p_token);
420 lwkt_reltoken(&p->p_token);
427 * Locate a process by number. The returned process will be referenced and
428 * must be released with PRELE().
435 struct proc *p = curproc;
439 * Shortcut the current process
441 if (p && p->p_pid == pid) {
447 * Otherwise find it in the hash table.
449 n = ALLPROC_HASH(pid);
451 lwkt_gettoken_shared(&proc_tokens[n]);
452 LIST_FOREACH(p, &allprocs[n], p_list) {
453 if (p->p_stat == SZOMB)
455 if (p->p_pid == pid) {
457 lwkt_reltoken(&proc_tokens[n]);
461 lwkt_reltoken(&proc_tokens[n]);
467 * Locate a process by number. The returned process is NOT referenced.
468 * The result will not be stable and is typically only used to validate
469 * against a process that the caller has in-hand.
476 struct proc *p = curproc;
480 * Shortcut the current process
482 if (p && p->p_pid == pid)
486 * Otherwise find it in the hash table.
488 n = ALLPROC_HASH(pid);
490 lwkt_gettoken_shared(&proc_tokens[n]);
491 LIST_FOREACH(p, &allprocs[n], p_list) {
492 if (p->p_stat == SZOMB)
494 if (p->p_pid == pid) {
495 lwkt_reltoken(&proc_tokens[n]);
499 lwkt_reltoken(&proc_tokens[n]);
505 * Locate a process on the zombie list. Return a process or NULL.
506 * The returned process will be referenced and the caller must release
509 * No other requirements.
514 struct proc *p = curproc;
518 * Shortcut the current process
520 if (p && p->p_pid == pid) {
526 * Otherwise find it in the hash table.
528 n = ALLPROC_HASH(pid);
530 lwkt_gettoken_shared(&proc_tokens[n]);
531 LIST_FOREACH(p, &allprocs[n], p_list) {
532 if (p->p_stat != SZOMB)
534 if (p->p_pid == pid) {
536 lwkt_reltoken(&proc_tokens[n]);
540 lwkt_reltoken(&proc_tokens[n]);
547 pgref(struct pgrp *pgrp)
549 refcount_acquire(&pgrp->pg_refs);
553 pgrel(struct pgrp *pgrp)
558 n = PGRP_HASH(pgrp->pg_id);
560 count = pgrp->pg_refs;
564 lwkt_gettoken(&proc_tokens[n]);
565 if (atomic_cmpset_int(&pgrp->pg_refs, 1, 0))
567 lwkt_reltoken(&proc_tokens[n]);
570 if (atomic_cmpset_int(&pgrp->pg_refs, count, count - 1))
577 * Successful 1->0 transition, pghash_spin is held.
579 LIST_REMOVE(pgrp, pg_list);
580 pid_doms[pgrp->pg_id % PIDSEL_DOMAINS] = (uint8_t)time_second;
583 * Reset any sigio structures pointing to us as a result of
584 * F_SETOWN with our pgid.
586 funsetownlst(&pgrp->pg_sigiolst);
588 if (pgrp->pg_session->s_ttyp != NULL &&
589 pgrp->pg_session->s_ttyp->t_pgrp == pgrp) {
590 pgrp->pg_session->s_ttyp->t_pgrp = NULL;
592 lwkt_reltoken(&proc_tokens[n]);
594 sess_rele(pgrp->pg_session);
599 * Locate a process group by number. The returned process group will be
600 * referenced w/pgref() and must be released with pgrel() (or assigned
601 * somewhere if you wish to keep the reference).
612 lwkt_gettoken_shared(&proc_tokens[n]);
614 LIST_FOREACH(pgrp, &allpgrps[n], pg_list) {
615 if (pgrp->pg_id == pgid) {
616 refcount_acquire(&pgrp->pg_refs);
617 lwkt_reltoken(&proc_tokens[n]);
621 lwkt_reltoken(&proc_tokens[n]);
626 * Move p to a new or existing process group (and session)
631 enterpgrp(struct proc *p, pid_t pgid, int mksess)
639 KASSERT(pgrp == NULL || !mksess,
640 ("enterpgrp: setsid into non-empty pgrp"));
641 KASSERT(!SESS_LEADER(p),
642 ("enterpgrp: session leader attempted setpgrp"));
645 pid_t savepid = p->p_pid;
652 KASSERT(p->p_pid == pgid,
653 ("enterpgrp: new pgrp and pid != pgid"));
654 pgrp = kmalloc(sizeof(struct pgrp), M_PGRP, M_WAITOK | M_ZERO);
656 LIST_INIT(&pgrp->pg_members);
658 SLIST_INIT(&pgrp->pg_sigiolst);
659 lwkt_token_init(&pgrp->pg_token, "pgrp_token");
660 refcount_init(&pgrp->pg_refs, 1);
661 lockinit(&pgrp->pg_lock, "pgwt", 0, 0);
665 if ((np = pfindn(savepid)) == NULL || np != p) {
666 lwkt_reltoken(&proc_tokens[n]);
672 lwkt_gettoken(&proc_tokens[n]);
674 struct session *sess;
679 sess = kmalloc(sizeof(struct session), M_SESSION,
681 lwkt_gettoken(&p->p_token);
683 sess->s_sid = p->p_pid;
685 sess->s_ttyvp = NULL;
687 bcopy(p->p_session->s_login, sess->s_login,
688 sizeof(sess->s_login));
689 pgrp->pg_session = sess;
690 KASSERT(p == curproc,
691 ("enterpgrp: mksession and p != curproc"));
692 p->p_flags &= ~P_CONTROLT;
693 LIST_INSERT_HEAD(&allsessn[n], sess, s_list);
694 lwkt_reltoken(&p->p_token);
696 lwkt_gettoken(&p->p_token);
697 pgrp->pg_session = p->p_session;
698 sess_hold(pgrp->pg_session);
699 lwkt_reltoken(&p->p_token);
701 LIST_INSERT_HEAD(&allpgrps[n], pgrp, pg_list);
703 lwkt_reltoken(&proc_tokens[n]);
704 } else if (pgrp == p->p_pgrp) {
707 } /* else pgfind() referenced the pgrp */
709 lwkt_gettoken(&pgrp->pg_token);
710 lwkt_gettoken(&p->p_token);
713 * Replace p->p_pgrp, handling any races that occur.
715 while ((opgrp = p->p_pgrp) != NULL) {
717 lwkt_gettoken(&opgrp->pg_token);
718 if (opgrp != p->p_pgrp) {
719 lwkt_reltoken(&opgrp->pg_token);
723 LIST_REMOVE(p, p_pglist);
727 LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist);
730 * Adjust eligibility of affected pgrps to participate in job control.
731 * Increment eligibility counts before decrementing, otherwise we
732 * could reach 0 spuriously during the first call.
736 fixjobc(p, opgrp, 0);
737 lwkt_reltoken(&opgrp->pg_token);
738 pgrel(opgrp); /* manual pgref */
739 pgrel(opgrp); /* p->p_pgrp ref */
741 lwkt_reltoken(&p->p_token);
742 lwkt_reltoken(&pgrp->pg_token);
750 * Remove process from process group
755 leavepgrp(struct proc *p)
757 struct pgrp *pg = p->p_pgrp;
759 lwkt_gettoken(&p->p_token);
760 while ((pg = p->p_pgrp) != NULL) {
762 lwkt_gettoken(&pg->pg_token);
763 if (p->p_pgrp != pg) {
764 lwkt_reltoken(&pg->pg_token);
769 LIST_REMOVE(p, p_pglist);
770 lwkt_reltoken(&pg->pg_token);
771 pgrel(pg); /* manual pgref */
772 pgrel(pg); /* p->p_pgrp ref */
775 lwkt_reltoken(&p->p_token);
781 * Adjust the ref count on a session structure. When the ref count falls to
782 * zero the tty is disassociated from the session and the session structure
783 * is freed. Note that tty assocation is not itself ref-counted.
788 sess_hold(struct session *sp)
790 atomic_add_int(&sp->s_count, 1);
797 sess_rele(struct session *sess)
803 n = SESS_HASH(sess->s_sid);
805 count = sess->s_count;
809 lwkt_gettoken(&tty_token);
810 lwkt_gettoken(&proc_tokens[n]);
811 if (atomic_cmpset_int(&sess->s_count, 1, 0))
813 lwkt_reltoken(&proc_tokens[n]);
814 lwkt_reltoken(&tty_token);
817 if (atomic_cmpset_int(&sess->s_count, count, count - 1))
824 * Successful 1->0 transition and tty_token is held.
826 LIST_REMOVE(sess, s_list);
827 pid_doms[sess->s_sid % PIDSEL_DOMAINS] = (uint8_t)time_second;
829 if (sess->s_ttyp && sess->s_ttyp->t_session) {
830 #ifdef TTY_DO_FULL_CLOSE
831 /* FULL CLOSE, see ttyclearsession() */
832 KKASSERT(sess->s_ttyp->t_session == sess);
833 sess->s_ttyp->t_session = NULL;
835 /* HALF CLOSE, see ttyclearsession() */
836 if (sess->s_ttyp->t_session == sess)
837 sess->s_ttyp->t_session = NULL;
840 if ((tp = sess->s_ttyp) != NULL) {
844 lwkt_reltoken(&proc_tokens[n]);
845 lwkt_reltoken(&tty_token);
847 kfree(sess, M_SESSION);
851 * Adjust pgrp jobc counters when specified process changes process group.
852 * We count the number of processes in each process group that "qualify"
853 * the group for terminal job control (those with a parent in a different
854 * process group of the same session). If that count reaches zero, the
855 * process group becomes orphaned. Check both the specified process'
856 * process group and that of its children.
857 * entering == 0 => p is leaving specified group.
858 * entering == 1 => p is entering specified group.
863 fixjobc(struct proc *p, struct pgrp *pgrp, int entering)
865 struct pgrp *hispgrp;
866 struct session *mysession;
870 * Check p's parent to see whether p qualifies its own process
871 * group; if so, adjust count for p's process group.
873 lwkt_gettoken(&p->p_token); /* p_children scan */
874 lwkt_gettoken(&pgrp->pg_token);
876 mysession = pgrp->pg_session;
877 if ((hispgrp = p->p_pptr->p_pgrp) != pgrp &&
878 hispgrp->pg_session == mysession) {
881 else if (--pgrp->pg_jobc == 0)
886 * Check this process' children to see whether they qualify
887 * their process groups; if so, adjust counts for children's
890 LIST_FOREACH(np, &p->p_children, p_sibling) {
892 lwkt_gettoken(&np->p_token);
893 if ((hispgrp = np->p_pgrp) != pgrp &&
894 hispgrp->pg_session == mysession &&
895 np->p_stat != SZOMB) {
897 lwkt_gettoken(&hispgrp->pg_token);
900 else if (--hispgrp->pg_jobc == 0)
902 lwkt_reltoken(&hispgrp->pg_token);
905 lwkt_reltoken(&np->p_token);
908 KKASSERT(pgrp->pg_refs > 0);
909 lwkt_reltoken(&pgrp->pg_token);
910 lwkt_reltoken(&p->p_token);
914 * A process group has become orphaned;
915 * if there are any stopped processes in the group,
916 * hang-up all process in that group.
918 * The caller must hold pg_token.
921 orphanpg(struct pgrp *pg)
925 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
926 if (p->p_stat == SSTOP) {
927 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
937 * Add a new process to the allproc list and the PID hash. This
938 * also assigns a pid to the new process.
943 proc_add_allproc(struct proc *p)
947 if ((random_offset = randompid) != 0) {
948 read_random(&random_offset, sizeof(random_offset));
949 random_offset = (random_offset & 0x7FFFFFFF) % randompid;
951 proc_makepid(p, random_offset);
955 * Calculate a new process pid. This function is integrated into
956 * proc_add_allproc() to guarentee that the new pid is not reused before
957 * the new process can be added to the allproc list.
959 * p_pid is assigned and the process is added to the allproc hash table
961 * WARNING! We need to allocate PIDs sequentially during early boot.
962 * In particular, init needs to have a pid of 1.
966 proc_makepid(struct proc *p, int random_offset)
968 static pid_t nextpid = 1; /* heuristic, allowed to race */
971 struct session *sess;
978 * Select the next pid base candidate.
980 * Check cyclement, do not allow a pid < 100.
984 base = atomic_fetchadd_int(&nextpid, 1) + random_offset;
985 if (base <= 0 || base >= PID_MAX) {
986 base = base % PID_MAX;
991 nextpid = base; /* reset (SMP race ok) */
995 * Do not allow a base pid to be selected from a domain that has
996 * recently seen a pid/pgid/sessid reap. Sleep a little if we looped
997 * through all available domains.
999 * WARNING: We want the early pids to be allocated linearly,
1000 * particularly pid 1 and pid 2.
1002 if (++retries >= PIDSEL_DOMAINS)
1003 tsleep(&nextpid, 0, "makepid", 1);
1005 delta8 = (int8_t)time_second -
1006 (int8_t)pid_doms[base % PIDSEL_DOMAINS];
1007 if (delta8 >= 0 && delta8 <= PIDDOM_DELAY) {
1014 * Calculate a hash index and find an unused process id within
1015 * the table, looping if we cannot find one.
1017 * The inner loop increments by ALLPROC_HSIZE which keeps the
1018 * PID at the same pid_doms[] index as well as the same hash index.
1020 n = ALLPROC_HASH(base);
1021 lwkt_gettoken(&proc_tokens[n]);
1024 LIST_FOREACH(ps, &allprocs[n], p_list) {
1025 if (ps->p_pid == base) {
1026 base += ALLPROC_HSIZE;
1027 if (base >= PID_MAX) {
1028 lwkt_reltoken(&proc_tokens[n]);
1035 LIST_FOREACH(pg, &allpgrps[n], pg_list) {
1036 if (pg->pg_id == base) {
1037 base += ALLPROC_HSIZE;
1038 if (base >= PID_MAX) {
1039 lwkt_reltoken(&proc_tokens[n]);
1046 LIST_FOREACH(sess, &allsessn[n], s_list) {
1047 if (sess->s_sid == base) {
1048 base += ALLPROC_HSIZE;
1049 if (base >= PID_MAX) {
1050 lwkt_reltoken(&proc_tokens[n]);
1059 * Assign the pid and insert the process.
1062 LIST_INSERT_HEAD(&allprocs[n], p, p_list);
1063 lwkt_reltoken(&proc_tokens[n]);
1067 * Called from exit1 to place the process into a zombie state.
1068 * The process is removed from the pid hash and p_stat is set
1069 * to SZOMB. Normal pfind[n]() calls will not find it any more.
1071 * Caller must hold p->p_token. We are required to wait until p_lock
1072 * becomes zero before we can manipulate the list, allowing allproc
1073 * scans to guarantee consistency during a list scan.
1076 proc_move_allproc_zombie(struct proc *p)
1080 n = ALLPROC_HASH(p->p_pid);
1081 PSTALL(p, "reap1", 0);
1082 lwkt_gettoken(&proc_tokens[n]);
1084 PSTALL(p, "reap1a", 0);
1087 lwkt_reltoken(&proc_tokens[n]);
1088 dsched_exit_proc(p);
1092 * This routine is called from kern_wait() and will remove the process
1093 * from the zombie list and the sibling list. This routine will block
1094 * if someone has a lock on the proces (p_lock).
1096 * Caller must hold p->p_token. We are required to wait until p_lock
1097 * becomes zero before we can manipulate the list, allowing allproc
1098 * scans to guarantee consistency during a list scan.
1101 proc_remove_zombie(struct proc *p)
1105 n = ALLPROC_HASH(p->p_pid);
1107 PSTALL(p, "reap2", 0);
1108 lwkt_gettoken(&proc_tokens[n]);
1109 PSTALL(p, "reap2a", 0);
1110 LIST_REMOVE(p, p_list); /* from remove master list */
1111 LIST_REMOVE(p, p_sibling); /* and from sibling list */
1113 pid_doms[p->p_pid % PIDSEL_DOMAINS] = (uint8_t)time_second;
1114 lwkt_reltoken(&proc_tokens[n]);
1118 * Handle various requirements prior to returning to usermode. Called from
1119 * platform trap and system call code.
1122 lwpuserret(struct lwp *lp)
1124 struct proc *p = lp->lwp_proc;
1126 if (lp->lwp_mpflags & LWP_MP_VNLRU) {
1127 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_VNLRU);
1130 if (lp->lwp_mpflags & LWP_MP_WEXIT) {
1131 lwkt_gettoken(&p->p_token);
1133 lwkt_reltoken(&p->p_token); /* NOT REACHED */
1138 * Kernel threads run from user processes can also accumulate deferred
1139 * actions which need to be acted upon. Callers include:
1141 * nfsd - Can allocate lots of vnodes
1144 lwpkthreaddeferred(void)
1146 struct lwp *lp = curthread->td_lwp;
1149 if (lp->lwp_mpflags & LWP_MP_VNLRU) {
1150 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_VNLRU);
1157 proc_usermap(struct proc *p, int invfork)
1159 struct sys_upmap *upmap;
1161 lwkt_gettoken(&p->p_token);
1162 upmap = kmalloc(roundup2(sizeof(*upmap), PAGE_SIZE), M_PROC,
1164 if (p->p_upmap == NULL) {
1165 upmap->header[0].type = UKPTYPE_VERSION;
1166 upmap->header[0].offset = offsetof(struct sys_upmap, version);
1167 upmap->header[1].type = UPTYPE_RUNTICKS;
1168 upmap->header[1].offset = offsetof(struct sys_upmap, runticks);
1169 upmap->header[2].type = UPTYPE_FORKID;
1170 upmap->header[2].offset = offsetof(struct sys_upmap, forkid);
1171 upmap->header[3].type = UPTYPE_PID;
1172 upmap->header[3].offset = offsetof(struct sys_upmap, pid);
1173 upmap->header[4].type = UPTYPE_PROC_TITLE;
1174 upmap->header[4].offset = offsetof(struct sys_upmap,proc_title);
1175 upmap->header[5].type = UPTYPE_INVFORK;
1176 upmap->header[5].offset = offsetof(struct sys_upmap, invfork);
1178 upmap->version = UPMAP_VERSION;
1179 upmap->pid = p->p_pid;
1180 upmap->forkid = p->p_forkid;
1181 upmap->invfork = invfork;
1184 kfree(upmap, M_PROC);
1186 lwkt_reltoken(&p->p_token);
1190 proc_userunmap(struct proc *p)
1192 struct sys_upmap *upmap;
1194 lwkt_gettoken(&p->p_token);
1195 if ((upmap = p->p_upmap) != NULL) {
1197 kfree(upmap, M_PROC);
1199 lwkt_reltoken(&p->p_token);
1203 * Scan all processes on the allproc list. The process is automatically
1204 * held for the callback. A return value of -1 terminates the loop.
1205 * Zombie procs are skipped.
1207 * The callback is made with the process held and proc_token held.
1209 * We limit the scan to the number of processes as-of the start of
1210 * the scan so as not to get caught up in an endless loop if new processes
1211 * are created more quickly than we can scan the old ones. Add a little
1212 * slop to try to catch edge cases since nprocs can race.
1217 allproc_scan(int (*callback)(struct proc *, void *), void *data)
1219 int limit = nprocs + ncpus;
1225 * proc_tokens[n] protects the allproc list and PHOLD() prevents the
1226 * process from being removed from the allproc list or the zombproc
1229 for (n = 0; n < ALLPROC_HSIZE; ++n) {
1230 if (LIST_FIRST(&allprocs[n]) == NULL)
1232 lwkt_gettoken(&proc_tokens[n]);
1233 LIST_FOREACH(p, &allprocs[n], p_list) {
1234 if (p->p_stat == SZOMB)
1237 r = callback(p, data);
1244 lwkt_reltoken(&proc_tokens[n]);
1247 * Check if asked to stop early
1255 * Scan all lwps of processes on the allproc list. The lwp is automatically
1256 * held for the callback. A return value of -1 terminates the loop.
1258 * The callback is made with the proces and lwp both held, and proc_token held.
1263 alllwp_scan(int (*callback)(struct lwp *, void *), void *data)
1270 for (n = 0; n < ALLPROC_HSIZE; ++n) {
1271 if (LIST_FIRST(&allprocs[n]) == NULL)
1273 lwkt_gettoken(&proc_tokens[n]);
1274 LIST_FOREACH(p, &allprocs[n], p_list) {
1275 if (p->p_stat == SZOMB)
1278 lwkt_gettoken(&p->p_token);
1279 FOREACH_LWP_IN_PROC(lp, p) {
1281 r = callback(lp, data);
1284 lwkt_reltoken(&p->p_token);
1289 lwkt_reltoken(&proc_tokens[n]);
1292 * Asked to exit early
1300 * Scan all processes on the zombproc list. The process is automatically
1301 * held for the callback. A return value of -1 terminates the loop.
1304 * The callback is made with the proces held and proc_token held.
1307 zombproc_scan(int (*callback)(struct proc *, void *), void *data)
1314 * proc_tokens[n] protects the allproc list and PHOLD() prevents the
1315 * process from being removed from the allproc list or the zombproc
1318 for (n = 0; n < ALLPROC_HSIZE; ++n) {
1319 if (LIST_FIRST(&allprocs[n]) == NULL)
1321 lwkt_gettoken(&proc_tokens[n]);
1322 LIST_FOREACH(p, &allprocs[n], p_list) {
1323 if (p->p_stat != SZOMB)
1326 r = callback(p, data);
1331 lwkt_reltoken(&proc_tokens[n]);
1334 * Check if asked to stop early
1341 #include "opt_ddb.h"
1343 #include <ddb/ddb.h>
1348 DB_SHOW_COMMAND(pgrpdump, pgrpdump)
1354 for (i = 0; i < ALLPROC_HSIZE; ++i) {
1355 if (LIST_EMPTY(&allpgrps[i]))
1357 kprintf("\tindx %d\n", i);
1358 LIST_FOREACH(pgrp, &allpgrps[i], pg_list) {
1359 kprintf("\tpgrp %p, pgid %ld, sess %p, "
1360 "sesscnt %d, mem %p\n",
1361 (void *)pgrp, (long)pgrp->pg_id,
1362 (void *)pgrp->pg_session,
1363 pgrp->pg_session->s_count,
1364 (void *)LIST_FIRST(&pgrp->pg_members));
1365 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1366 kprintf("\t\tpid %ld addr %p pgrp %p\n",
1367 (long)p->p_pid, (void *)p,
1376 * The caller must hold proc_token.
1379 sysctl_out_proc(struct proc *p, struct sysctl_req *req, int flags)
1381 struct kinfo_proc ki;
1383 int skp = 0, had_output = 0;
1386 bzero(&ki, sizeof(ki));
1387 lwkt_gettoken_shared(&p->p_token);
1388 fill_kinfo_proc(p, &ki);
1389 if ((flags & KERN_PROC_FLAG_LWP) == 0)
1392 FOREACH_LWP_IN_PROC(lp, p) {
1394 fill_kinfo_lwp(lp, &ki.kp_lwp);
1396 error = SYSCTL_OUT(req, &ki, sizeof(ki));
1403 lwkt_reltoken(&p->p_token);
1404 /* We need to output at least the proc, even if there is no lwp. */
1405 if (had_output == 0) {
1406 error = SYSCTL_OUT(req, &ki, sizeof(ki));
1412 * The caller must hold proc_token.
1415 sysctl_out_proc_kthread(struct thread *td, struct sysctl_req *req)
1417 struct kinfo_proc ki;
1420 fill_kinfo_proc_kthread(td, &ki);
1421 error = SYSCTL_OUT(req, &ki, sizeof(ki));
1431 sysctl_kern_proc(SYSCTL_HANDLER_ARGS)
1433 int *name = (int *)arg1;
1434 int oid = oidp->oid_number;
1435 u_int namelen = arg2;
1438 struct thread *marker;
1443 struct ucred *cr1 = curproc->p_ucred;
1445 flags = oid & KERN_PROC_FLAGMASK;
1446 oid &= ~KERN_PROC_FLAGMASK;
1448 if ((oid == KERN_PROC_ALL && namelen != 0) ||
1449 (oid != KERN_PROC_ALL && namelen != 1)) {
1454 * proc_token protects the allproc list and PHOLD() prevents the
1455 * process from being removed from the allproc list or the zombproc
1458 if (oid == KERN_PROC_PID) {
1459 p = pfind((pid_t)name[0]);
1461 if (PRISON_CHECK(cr1, p->p_ucred))
1462 error = sysctl_out_proc(p, req, flags);
1470 /* overestimate by 5 procs */
1471 error = SYSCTL_OUT(req, 0, sizeof (struct kinfo_proc) * 5);
1476 for (n = 0; n < ALLPROC_HSIZE; ++n) {
1477 if (LIST_EMPTY(&allprocs[n]))
1479 lwkt_gettoken_shared(&proc_tokens[n]);
1480 LIST_FOREACH(p, &allprocs[n], p_list) {
1482 * Show a user only their processes.
1484 if ((!ps_showallprocs) &&
1485 (p->p_ucred == NULL || p_trespass(cr1, p->p_ucred))) {
1489 * Skip embryonic processes.
1491 if (p->p_stat == SIDL)
1494 * TODO - make more efficient (see notes below).
1498 case KERN_PROC_PGRP:
1499 /* could do this by traversing pgrp */
1500 if (p->p_pgrp == NULL ||
1501 p->p_pgrp->pg_id != (pid_t)name[0])
1506 if ((p->p_flags & P_CONTROLT) == 0 ||
1507 p->p_session == NULL ||
1508 p->p_session->s_ttyp == NULL ||
1509 dev2udev(p->p_session->s_ttyp->t_dev) !=
1515 if (p->p_ucred == NULL ||
1516 p->p_ucred->cr_uid != (uid_t)name[0])
1520 case KERN_PROC_RUID:
1521 if (p->p_ucred == NULL ||
1522 p->p_ucred->cr_ruid != (uid_t)name[0])
1527 if (!PRISON_CHECK(cr1, p->p_ucred))
1530 error = sysctl_out_proc(p, req, flags);
1533 lwkt_reltoken(&proc_tokens[n]);
1537 lwkt_reltoken(&proc_tokens[n]);
1541 * Iterate over all active cpus and scan their thread list. Start
1542 * with the next logical cpu and end with our original cpu. We
1543 * migrate our own thread to each target cpu in order to safely scan
1544 * its thread list. In the last loop we migrate back to our original
1547 origcpu = mycpu->gd_cpuid;
1548 if (!ps_showallthreads || jailed(cr1))
1551 marker = kmalloc(sizeof(struct thread), M_TEMP, M_WAITOK|M_ZERO);
1552 marker->td_flags = TDF_MARKER;
1555 for (n = 1; n <= ncpus; ++n) {
1559 nid = (origcpu + n) % ncpus;
1560 if (CPUMASK_TESTBIT(smp_active_mask, nid) == 0)
1562 rgd = globaldata_find(nid);
1563 lwkt_setcpu_self(rgd);
1566 TAILQ_INSERT_TAIL(&rgd->gd_tdallq, marker, td_allq);
1568 while ((td = TAILQ_PREV(marker, lwkt_queue, td_allq)) != NULL) {
1569 TAILQ_REMOVE(&rgd->gd_tdallq, marker, td_allq);
1570 TAILQ_INSERT_BEFORE(td, marker, td_allq);
1571 if (td->td_flags & TDF_MARKER)
1580 case KERN_PROC_PGRP:
1583 case KERN_PROC_RUID:
1586 error = sysctl_out_proc_kthread(td, req);
1594 TAILQ_REMOVE(&rgd->gd_tdallq, marker, td_allq);
1602 * Userland scheduler expects us to return on the same cpu we
1605 if (mycpu->gd_cpuid != origcpu)
1606 lwkt_setcpu_self(globaldata_find(origcpu));
1608 kfree(marker, M_TEMP);
1615 * This sysctl allows a process to retrieve the argument list or process
1616 * title for another process without groping around in the address space
1617 * of the other process. It also allow a process to set its own "process
1618 * title to a string of its own choice.
1623 sysctl_kern_proc_args(SYSCTL_HANDLER_ARGS)
1625 int *name = (int*) arg1;
1626 u_int namelen = arg2;
1631 struct ucred *cr1 = curproc->p_ucred;
1636 p = pfind((pid_t)name[0]);
1639 lwkt_gettoken(&p->p_token);
1641 if ((!ps_argsopen) && p_trespass(cr1, p->p_ucred))
1644 if (req->newptr && curproc != p) {
1649 if (p->p_upmap != NULL && p->p_upmap->proc_title[0]) {
1651 * Args set via writable user process mmap.
1652 * We must calculate the string length manually
1653 * because the user data can change at any time.
1658 base = p->p_upmap->proc_title;
1659 for (n = 0; n < UPMAP_MAXPROCTITLE - 1; ++n) {
1663 error = SYSCTL_OUT(req, base, n);
1665 error = SYSCTL_OUT(req, "", 1);
1666 } else if ((pa = p->p_args) != NULL) {
1668 * Args set by setproctitle() sysctl.
1670 refcount_acquire(&pa->ar_ref);
1671 error = SYSCTL_OUT(req, pa->ar_args, pa->ar_length);
1672 if (refcount_release(&pa->ar_ref))
1676 if (req->newptr == NULL)
1679 if (req->newlen + sizeof(struct pargs) > ps_arg_cache_limit) {
1683 pa = kmalloc(sizeof(struct pargs) + req->newlen, M_PARGS, M_WAITOK);
1684 refcount_init(&pa->ar_ref, 1);
1685 pa->ar_length = req->newlen;
1686 error = SYSCTL_IN(req, pa->ar_args, req->newlen);
1694 * Replace p_args with the new pa. p_args may have previously
1701 KKASSERT(opa->ar_ref > 0);
1702 if (refcount_release(&opa->ar_ref)) {
1703 kfree(opa, M_PARGS);
1709 lwkt_reltoken(&p->p_token);
1716 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
1718 int *name = (int*) arg1;
1719 u_int namelen = arg2;
1722 char *fullpath, *freepath;
1723 struct ucred *cr1 = curproc->p_ucred;
1728 p = pfind((pid_t)name[0]);
1731 lwkt_gettoken_shared(&p->p_token);
1734 * If we are not allowed to see other args, we certainly shouldn't
1735 * get the cwd either. Also check the usual trespassing.
1737 if ((!ps_argsopen) && p_trespass(cr1, p->p_ucred))
1740 if (req->oldptr && p->p_fd != NULL && p->p_fd->fd_ncdir.ncp) {
1741 struct nchandle nch;
1743 cache_copy(&p->p_fd->fd_ncdir, &nch);
1744 error = cache_fullpath(p, &nch, NULL,
1745 &fullpath, &freepath, 0);
1749 error = SYSCTL_OUT(req, fullpath, strlen(fullpath) + 1);
1750 kfree(freepath, M_TEMP);
1755 lwkt_reltoken(&p->p_token);
1762 * This sysctl allows a process to retrieve the path of the executable for
1763 * itself or another process.
1766 sysctl_kern_proc_pathname(SYSCTL_HANDLER_ARGS)
1768 pid_t *pidp = (pid_t *)arg1;
1769 unsigned int arglen = arg2;
1772 char *retbuf, *freebuf;
1777 if (*pidp == -1) { /* -1 means this process */
1790 error = vn_fullpath(p, vp, &retbuf, &freebuf, 0);
1794 error = SYSCTL_OUT(req, retbuf, strlen(retbuf) + 1);
1795 kfree(freebuf, M_TEMP);
1803 SYSCTL_NODE(_kern, KERN_PROC, proc, CTLFLAG_RD, 0, "Process table");
1805 SYSCTL_PROC(_kern_proc, KERN_PROC_ALL, all, CTLFLAG_RD|CTLTYPE_STRUCT,
1806 0, 0, sysctl_kern_proc, "S,proc", "Return entire process table");
1808 SYSCTL_NODE(_kern_proc, KERN_PROC_PGRP, pgrp, CTLFLAG_RD,
1809 sysctl_kern_proc, "Process table");
1811 SYSCTL_NODE(_kern_proc, KERN_PROC_TTY, tty, CTLFLAG_RD,
1812 sysctl_kern_proc, "Process table");
1814 SYSCTL_NODE(_kern_proc, KERN_PROC_UID, uid, CTLFLAG_RD,
1815 sysctl_kern_proc, "Process table");
1817 SYSCTL_NODE(_kern_proc, KERN_PROC_RUID, ruid, CTLFLAG_RD,
1818 sysctl_kern_proc, "Process table");
1820 SYSCTL_NODE(_kern_proc, KERN_PROC_PID, pid, CTLFLAG_RD,
1821 sysctl_kern_proc, "Process table");
1823 SYSCTL_NODE(_kern_proc, (KERN_PROC_ALL | KERN_PROC_FLAG_LWP), all_lwp, CTLFLAG_RD,
1824 sysctl_kern_proc, "Process table");
1826 SYSCTL_NODE(_kern_proc, (KERN_PROC_PGRP | KERN_PROC_FLAG_LWP), pgrp_lwp, CTLFLAG_RD,
1827 sysctl_kern_proc, "Process table");
1829 SYSCTL_NODE(_kern_proc, (KERN_PROC_TTY | KERN_PROC_FLAG_LWP), tty_lwp, CTLFLAG_RD,
1830 sysctl_kern_proc, "Process table");
1832 SYSCTL_NODE(_kern_proc, (KERN_PROC_UID | KERN_PROC_FLAG_LWP), uid_lwp, CTLFLAG_RD,
1833 sysctl_kern_proc, "Process table");
1835 SYSCTL_NODE(_kern_proc, (KERN_PROC_RUID | KERN_PROC_FLAG_LWP), ruid_lwp, CTLFLAG_RD,
1836 sysctl_kern_proc, "Process table");
1838 SYSCTL_NODE(_kern_proc, (KERN_PROC_PID | KERN_PROC_FLAG_LWP), pid_lwp, CTLFLAG_RD,
1839 sysctl_kern_proc, "Process table");
1841 SYSCTL_NODE(_kern_proc, KERN_PROC_ARGS, args, CTLFLAG_RW | CTLFLAG_ANYBODY,
1842 sysctl_kern_proc_args, "Process argument list");
1844 SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD | CTLFLAG_ANYBODY,
1845 sysctl_kern_proc_cwd, "Process argument list");
1847 static SYSCTL_NODE(_kern_proc, KERN_PROC_PATHNAME, pathname, CTLFLAG_RD,
1848 sysctl_kern_proc_pathname, "Process executable path");