xref: /dragonfly/sys/kern/kern_proc.c (revision ec1c3f3acbcc9eec2e3e83bf248a95cca926a503)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *        The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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.
16  *
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
27  * SUCH DAMAGE.
28  */
29 
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>
35 #include <sys/proc.h>
36 #include <sys/vnode.h>
37 #include <sys/jail.h>
38 #include <sys/filedesc.h>
39 #include <sys/tty.h>
40 #include <sys/dsched.h>
41 #include <sys/signalvar.h>
42 #include <sys/spinlock.h>
43 #include <sys/random.h>
44 #include <sys/exec.h>
45 #include <vm/vm.h>
46 #include <sys/lock.h>
47 #include <sys/kinfo.h>
48 #include <vm/pmap.h>
49 #include <vm/vm_map.h>
50 #include <machine/smp.h>
51 
52 #include <sys/refcount.h>
53 #include <sys/spinlock2.h>
54 
55 /*
56  * Hash table size must be a power of two and is not currently dynamically
57  * sized.  There is a trade-off between the linear scans which must iterate
58  * all HSIZE elements and the number of elements which might accumulate
59  * within each hash chain.
60  */
61 #define ALLPROC_HSIZE         256
62 #define ALLPROC_HMASK         (ALLPROC_HSIZE - 1)
63 #define ALLPROC_HASH(pid)     (pid & ALLPROC_HMASK)
64 #define PGRP_HASH(pid)        (pid & ALLPROC_HMASK)
65 #define SESS_HASH(pid)        (pid & ALLPROC_HMASK)
66 
67 /*
68  * pid_doms[] management, used to control how quickly a PID can be recycled.
69  * Must be a multiple of ALLPROC_HSIZE for the proc_makepid() inner loops.
70  *
71  * WARNING! PIDDOM_DELAY should not be defined > 20 or so unless you change
72  *            the array from int8_t's to int16_t's.
73  */
74 #define PIDDOM_COUNT          10        /* 10 pids per domain - reduce array size */
75 #define PIDDOM_DELAY          10        /* min 10 seconds after exit before reuse */
76 #define PIDDOM_SCALE          10        /* (10,000*SCALE)/sec performance guarantee */
77 #define PIDSEL_DOMAINS        rounddown(PID_MAX * PIDDOM_SCALE / PIDDOM_COUNT, ALLPROC_HSIZE)
78 
79 /* Used by libkvm */
80 int allproc_hsize = ALLPROC_HSIZE;
81 
82 LIST_HEAD(pidhashhead, proc);
83 
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");
89 MALLOC_DEFINE(M_UPMAP, "upmap", "upmap/kpmap/lpmap structures");
90 
91 int ps_showallprocs = 1;
92 static int ps_showallthreads = 1;
93 SYSCTL_INT(_security, OID_AUTO, ps_showallprocs, CTLFLAG_RW,
94     &ps_showallprocs, 0,
95     "Unprivileged processes can see processes with different UID/GID");
96 SYSCTL_INT(_security, OID_AUTO, ps_showallthreads, CTLFLAG_RW,
97     &ps_showallthreads, 0,
98     "Unprivileged processes can see kernel threads");
99 static u_int pid_domain_skips;
100 SYSCTL_UINT(_kern, OID_AUTO, pid_domain_skips, CTLFLAG_RW,
101     &pid_domain_skips, 0,
102     "Number of pid_doms[] skipped");
103 static u_int pid_inner_skips;
104 SYSCTL_UINT(_kern, OID_AUTO, pid_inner_skips, CTLFLAG_RW,
105     &pid_inner_skips, 0,
106     "Number of pid_doms[] skipped");
107 
108 static void orphanpg(struct pgrp *pg);
109 static void proc_makepid(struct proc *p, int random_offset);
110 
111 /*
112  * Process related lists (for proc_token, allproc, allpgrp, and allsess)
113  */
114 typedef struct procglob procglob_t;
115 
116 static procglob_t   procglob[ALLPROC_HSIZE];
117 
118 /*
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.
122  *
123  * This array has to be fairly large to support a high fork/exec rate.
124  * A ~100,000 entry array will support a 10-second reuse latency at
125  * 10,000 execs/second, worst case.  Best-case multiply by PIDDOM_COUNT
126  * (approximately 100,000 execs/second).
127  *
128  * Currently we allocate around a megabyte, making the worst-case fork
129  * rate around 100,000/second.
130  */
131 static uint8_t *pid_doms;
132 
133 /*
134  * Random component to nextpid generation.  We mix in a random factor to make
135  * it a little harder to predict.  We sanity check the modulus value to avoid
136  * doing it in critical paths.  Don't let it be too small or we pointlessly
137  * waste randomness entropy, and don't let it be impossibly large.  Using a
138  * modulus that is too big causes a LOT more process table scans and slows
139  * down fork processing as the pidchecked caching is defeated.
140  */
141 static int randompid = 0;
142 
143 static __inline
144 struct ucred *
pcredcache(struct ucred * cr,struct proc * p)145 pcredcache(struct ucred *cr, struct proc *p)
146 {
147           if (cr != p->p_ucred) {
148                     if (cr)
149                               crfree(cr);
150                     spin_lock(&p->p_spin);
151                     if ((cr = p->p_ucred) != NULL)
152                               crhold(cr);
153                     spin_unlock(&p->p_spin);
154           }
155           return cr;
156 }
157 
158 /*
159  * No requirements.
160  */
161 static int
sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)162 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS)
163 {
164           int error, pid;
165 
166           pid = randompid;
167           error = sysctl_handle_int(oidp, &pid, 0, req);
168           if (error || !req->newptr)
169                     return (error);
170           if (pid < 0 || pid > PID_MAX - 100)     /* out of range */
171                     pid = PID_MAX - 100;
172           else if (pid < 2)                       /* NOP */
173                     pid = 0;
174           else if (pid < 100)                     /* Make it reasonable */
175                     pid = 100;
176           randompid = pid;
177           return (error);
178 }
179 
180 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW,
181               0, 0, sysctl_kern_randompid, "I", "Random PID modulus");
182 
183 /*
184  * Initialize global process hashing structures.
185  *
186  * These functions are ONLY called from the low level boot code and do
187  * not lock their operations.
188  */
189 void
procinit(void)190 procinit(void)
191 {
192           u_long i;
193 
194           /*
195            * Allocate dynamically.  This array can be large (~1MB) so don't
196            * waste boot loader space.
197            */
198           pid_doms = kmalloc(sizeof(pid_doms[0]) * PIDSEL_DOMAINS,
199                                  M_PROC, M_WAITOK | M_ZERO);
200 
201           /*
202            * Avoid unnecessary stalls due to pid_doms[] values all being
203            * the same.  Make sure that the allocation of pid 1 and pid 2
204            * succeeds.
205            */
206           for (i = 0; i < PIDSEL_DOMAINS; ++i)
207                     pid_doms[i] = (int8_t)i - (int8_t)(PIDDOM_DELAY + 1);
208 
209           /*
210            * Other misc init.
211            */
212           for (i = 0; i < ALLPROC_HSIZE; ++i) {
213                     procglob_t *prg = &procglob[i];
214                     LIST_INIT(&prg->allproc);
215                     LIST_INIT(&prg->allsess);
216                     LIST_INIT(&prg->allpgrp);
217                     lwkt_token_init(&prg->proc_token, "allproc");
218           }
219           uihashinit();
220 }
221 
222 void
procinsertinit(struct proc * p)223 procinsertinit(struct proc *p)
224 {
225           LIST_INSERT_HEAD(&procglob[ALLPROC_HASH(p->p_pid)].allproc,
226                                p, p_list);
227 }
228 
229 void
pgrpinsertinit(struct pgrp * pg)230 pgrpinsertinit(struct pgrp *pg)
231 {
232           LIST_INSERT_HEAD(&procglob[ALLPROC_HASH(pg->pg_id)].allpgrp,
233                                pg, pg_list);
234 }
235 
236 void
sessinsertinit(struct session * sess)237 sessinsertinit(struct session *sess)
238 {
239           LIST_INSERT_HEAD(&procglob[ALLPROC_HASH(sess->s_sid)].allsess,
240                                sess, s_list);
241 }
242 
243 /*
244  * Process hold/release support functions.  Called via the PHOLD(),
245  * PRELE(), and PSTALL() macros.
246  *
247  * p->p_lock is a simple hold count with a waiting interlock.  No wakeup()
248  * is issued unless someone is actually waiting for the process.
249  *
250  * Most holds are short-term, allowing a process scan or other similar
251  * operation to access a proc structure without it getting ripped out from
252  * under us.  procfs and process-list sysctl ops also use the hold function
253  * interlocked with various p_flags to keep the vmspace intact when reading
254  * or writing a user process's address space.
255  *
256  * There are two situations where a hold count can be longer.  Exiting lwps
257  * hold the process until the lwp is reaped, and the parent will hold the
258  * child during vfork()/exec() sequences while the child is marked P_PPWAIT.
259  *
260  * The kernel waits for the hold count to drop to 0 (or 1 in some cases) at
261  * various critical points in the fork/exec and exit paths before proceeding.
262  */
263 #define PLOCK_WAITING         0x40000000          /* tsleep() on p_lock */
264 #define PLOCK_ZOMB  0x20000000          /* zombie interlock held */
265 #define PLOCK_WAITRES         0x10000000          /* wait reservation held */
266 #define PLOCK_MASK  0x0FFFFFFF
267 
268 /*
269  * Returns non-zero if the WAITRES flag has been set
270  */
271 int
pwaitres_pending(struct proc * p)272 pwaitres_pending(struct proc *p)
273 {
274           if (p->p_lock & PLOCK_WAITRES)
275                     return 1;
276           return 0;
277 }
278 
279 /*
280  * Caller holds PLOCK_ZOMB.  Sets PLOCK_WAITRES and wakes up anyone in
281  * pholdzomb() (which will fail).
282  */
283 void
pwaitres_set(struct proc * p)284 pwaitres_set(struct proc *p)
285 {
286           int o;
287 
288           KKASSERT((p->p_lock & (PLOCK_ZOMB | PLOCK_WAITRES)) == PLOCK_ZOMB);
289           o = p->p_lock;
290           cpu_ccfence();
291           for (;;) {
292                     if (atomic_fcmpset_int(&p->p_lock, &o,
293                                                (o | PLOCK_WAITRES) & ~PLOCK_WAITING)) {
294                               if (o & PLOCK_WAITING)
295                                         wakeup(&p->p_lock);
296                               return;
297                     }
298           }
299 }
300 
301 void
pstall(struct proc * p,const char * wmesg,int count)302 pstall(struct proc *p, const char *wmesg, int count)
303 {
304           int o;
305           int n;
306 
307           for (;;) {
308                     o = p->p_lock;
309                     cpu_ccfence();
310                     if ((o & PLOCK_MASK) <= count)
311                               break;
312                     n = o | PLOCK_WAITING;
313                     tsleep_interlock(&p->p_lock, 0);
314 
315                     /*
316                      * If someone is trying to single-step the process during
317                      * an exec or an exit they can deadlock us because procfs
318                      * sleeps with the process held.
319                      */
320                     if (p->p_stops) {
321                               if (p->p_flags & P_INEXEC) {
322                                         wakeup(&p->p_stype);
323                               } else if (p->p_flags & P_POSTEXIT) {
324                                         spin_lock(&p->p_spin);
325                                         p->p_stops = 0;
326                                         p->p_step = 0;
327                                         spin_unlock(&p->p_spin);
328                                         wakeup(&p->p_stype);
329                               }
330                     }
331 
332                     if (atomic_cmpset_int(&p->p_lock, o, n)) {
333                               tsleep(&p->p_lock, PINTERLOCKED, wmesg, 0);
334                     }
335           }
336 }
337 
338 void
phold(struct proc * p)339 phold(struct proc *p)
340 {
341           atomic_add_int(&p->p_lock, 1);
342 }
343 
344 /*
345  * WARNING!  On last release (p) can become instantly invalid due to
346  *             MP races.
347  */
348 void
prele(struct proc * p)349 prele(struct proc *p)
350 {
351           int o;
352           int n;
353 
354           /*
355            * Fast path
356            */
357           if (atomic_cmpset_int(&p->p_lock, 1, 0))
358                     return;
359 
360           /*
361            * Slow path
362            */
363           for (;;) {
364                     o = p->p_lock;
365                     KKASSERT((o & PLOCK_MASK) > 0);
366                     cpu_ccfence();
367                     n = (o - 1) & ~PLOCK_WAITING;
368                     if (atomic_cmpset_int(&p->p_lock, o, n)) {
369                               if (o & PLOCK_WAITING)
370                                         wakeup(&p->p_lock);
371                               break;
372                     }
373           }
374 }
375 
376 /*
377  * Hold and flag serialized for zombie reaping purposes.  Fail if we had
378  * to sleep or if another thread has reserved the reap (WAITRES).
379  *
380  * This function will fail if it has to block, returning non-zero with
381  * neither the flag set or the hold count bumped.  Note that (p) may
382  * not be valid in this case if the caller does not have some other
383  * reference on (p).
384  *
385  * This function does not block on other PHOLD()s, only on other
386  * PHOLDZOMB()s.
387  *
388  * Zero is returned on success.  The hold count will be incremented and
389  * the serialization flag acquired.  Note that serialization is only against
390  * other pholdzomb() calls, not against phold() calls.
391  */
392 int
pholdzomb(struct proc * p)393 pholdzomb(struct proc *p)
394 {
395           int o;
396           int n;
397 
398           /*
399            * Fast path
400            */
401           if (atomic_cmpset_int(&p->p_lock, 0, PLOCK_ZOMB | 1))
402                     return(0);
403 
404           /*
405            * Slow path
406            */
407           for (;;) {
408                     o = p->p_lock;
409                     cpu_ccfence();
410                     if ((o & (PLOCK_ZOMB | PLOCK_WAITRES)) == 0) {
411                               n = (o + 1) | PLOCK_ZOMB;
412                               if (atomic_cmpset_int(&p->p_lock, o, n))
413                                         return(0);
414                     } else if (o & PLOCK_WAITRES) {
415                               return(1);
416                     } else {
417                               KKASSERT((o & PLOCK_MASK) > 0);
418                               n = o | PLOCK_WAITING;
419                               tsleep_interlock(&p->p_lock, 0);
420                               if (atomic_cmpset_int(&p->p_lock, o, n)) {
421                                         tsleep(&p->p_lock, PINTERLOCKED, "phldz", 0);
422                                         /* (p) can be ripped out at this point */
423                                         return(1);
424                               }
425                     }
426           }
427 }
428 
429 /*
430  * Release PLOCK_ZOMB, PLOCK_WAITRES, and the hold count, waking up any
431  * waiters.
432  *
433  * WARNING!  On last release (p) can become instantly invalid due to
434  *             MP races.
435  */
436 void
prelezomb(struct proc * p)437 prelezomb(struct proc *p)
438 {
439           int o;
440           int n;
441 
442           /*
443            * Fast path
444            */
445           if (atomic_cmpset_int(&p->p_lock, PLOCK_ZOMB | 1, 0))
446                     return;
447 
448           /*
449            * Slow path
450            */
451           KKASSERT(p->p_lock & PLOCK_ZOMB);
452           for (;;) {
453                     o = p->p_lock;
454                     KKASSERT((o & PLOCK_MASK) > 0);
455                     cpu_ccfence();
456                     n = (o - 1) & ~(PLOCK_ZOMB | PLOCK_WAITING | PLOCK_WAITRES);
457                     if (atomic_cmpset_int(&p->p_lock, o, n)) {
458                               if (o & PLOCK_WAITING)
459                                         wakeup(&p->p_lock);
460                               break;
461                     }
462           }
463 }
464 
465 /*
466  * Is p an inferior of the current process?
467  *
468  * No requirements.
469  */
470 int
inferior(struct proc * p)471 inferior(struct proc *p)
472 {
473           struct proc *p2;
474 
475           PHOLD(p);
476           lwkt_gettoken_shared(&p->p_token);
477           while (p != curproc) {
478                     if (p->p_pid == 0) {
479                               lwkt_reltoken(&p->p_token);
480                               return (0);
481                     }
482                     p2 = p->p_pptr;
483                     PHOLD(p2);
484                     lwkt_reltoken(&p->p_token);
485                     PRELE(p);
486                     lwkt_gettoken_shared(&p2->p_token);
487                     p = p2;
488           }
489           lwkt_reltoken(&p->p_token);
490           PRELE(p);
491 
492           return (1);
493 }
494 
495 /*
496  * Locate a process by number.  The returned process will be referenced and
497  * must be released with PRELE().
498  *
499  * No requirements.
500  */
501 struct proc *
pfind(pid_t pid)502 pfind(pid_t pid)
503 {
504           struct proc *p = curproc;
505           procglob_t *prg;
506           int n;
507 
508           /*
509            * Shortcut the current process
510            */
511           if (p && p->p_pid == pid) {
512                     PHOLD(p);
513                     return (p);
514           }
515 
516           /*
517            * Otherwise find it in the hash table.
518            */
519           n = ALLPROC_HASH(pid);
520           prg = &procglob[n];
521 
522           lwkt_gettoken_shared(&prg->proc_token);
523           LIST_FOREACH(p, &prg->allproc, p_list) {
524                     if (p->p_stat == SZOMB)
525                               continue;
526                     if (p->p_pid == pid) {
527                               PHOLD(p);
528                               lwkt_reltoken(&prg->proc_token);
529                               return (p);
530                     }
531           }
532           lwkt_reltoken(&prg->proc_token);
533 
534           return (NULL);
535 }
536 
537 /*
538  * Locate a process by number.  The returned process is NOT referenced.
539  * The result will not be stable and is typically only used to validate
540  * against a process that the caller has in-hand.
541  *
542  * No requirements.
543  */
544 struct proc *
pfindn(pid_t pid)545 pfindn(pid_t pid)
546 {
547           struct proc *p = curproc;
548           procglob_t *prg;
549           int n;
550 
551           /*
552            * Shortcut the current process
553            */
554           if (p && p->p_pid == pid)
555                     return (p);
556 
557           /*
558            * Otherwise find it in the hash table.
559            */
560           n = ALLPROC_HASH(pid);
561           prg = &procglob[n];
562 
563           lwkt_gettoken_shared(&prg->proc_token);
564           LIST_FOREACH(p, &prg->allproc, p_list) {
565                     if (p->p_stat == SZOMB)
566                               continue;
567                     if (p->p_pid == pid) {
568                               lwkt_reltoken(&prg->proc_token);
569                               return (p);
570                     }
571           }
572           lwkt_reltoken(&prg->proc_token);
573 
574           return (NULL);
575 }
576 
577 /*
578  * Locate a process on the zombie list.  Return a process or NULL.
579  * The returned process will be referenced and the caller must release
580  * it with PRELE().
581  *
582  * No other requirements.
583  */
584 struct proc *
zpfind(pid_t pid)585 zpfind(pid_t pid)
586 {
587           struct proc *p = curproc;
588           procglob_t *prg;
589           int n;
590 
591           /*
592            * Shortcut the current process
593            */
594           if (p && p->p_pid == pid) {
595                     PHOLD(p);
596                     return (p);
597           }
598 
599           /*
600            * Otherwise find it in the hash table.
601            */
602           n = ALLPROC_HASH(pid);
603           prg = &procglob[n];
604 
605           lwkt_gettoken_shared(&prg->proc_token);
606           LIST_FOREACH(p, &prg->allproc, p_list) {
607                     if (p->p_stat != SZOMB)
608                               continue;
609                     if (p->p_pid == pid) {
610                               PHOLD(p);
611                               lwkt_reltoken(&prg->proc_token);
612                               return (p);
613                     }
614           }
615           lwkt_reltoken(&prg->proc_token);
616 
617           return (NULL);
618 }
619 
620 /*
621  * Caller must hold the process token shared or exclusive.
622  * The returned lwp, if not NULL, will be held.  Caller must
623  * LWPRELE() it when done.
624  */
625 struct lwp *
lwpfind(struct proc * p,lwpid_t tid)626 lwpfind(struct proc *p, lwpid_t tid)
627 {
628           struct lwp *lp;
629 
630           lp = lwp_rb_tree_RB_LOOKUP(&p->p_lwp_tree, tid);
631           if (lp)
632                     LWPHOLD(lp);
633           return lp;
634 }
635 
636 void
pgref(struct pgrp * pgrp)637 pgref(struct pgrp *pgrp)
638 {
639           refcount_acquire(&pgrp->pg_refs);
640 }
641 
642 void
pgrel(struct pgrp * pgrp)643 pgrel(struct pgrp *pgrp)
644 {
645           procglob_t *prg;
646           int count;
647           int n;
648 
649           n = PGRP_HASH(pgrp->pg_id);
650           prg = &procglob[n];
651 
652           for (;;) {
653                     count = pgrp->pg_refs;
654                     cpu_ccfence();
655                     KKASSERT(count > 0);
656                     if (count == 1) {
657                               lwkt_gettoken(&prg->proc_token);
658                               if (atomic_cmpset_int(&pgrp->pg_refs, 1, 0))
659                                         break;
660                               lwkt_reltoken(&prg->proc_token);
661                               /* retry */
662                     } else {
663                               if (atomic_cmpset_int(&pgrp->pg_refs, count, count - 1))
664                                         return;
665                               /* retry */
666                     }
667           }
668 
669           /*
670            * Successful 1->0 transition, pghash_spin is held.
671            */
672           LIST_REMOVE(pgrp, pg_list);
673           if (pid_doms[pgrp->pg_id % PIDSEL_DOMAINS] != (uint8_t)time_second)
674                     pid_doms[pgrp->pg_id % PIDSEL_DOMAINS] = (uint8_t)time_second;
675 
676           /*
677            * Reset any sigio structures pointing to us as a result of
678            * F_SETOWN with our pgid.
679            */
680           funsetownlst(&pgrp->pg_sigiolst);
681 
682           if (pgrp->pg_session->s_ttyp != NULL &&
683               pgrp->pg_session->s_ttyp->t_pgrp == pgrp) {
684                     pgrp->pg_session->s_ttyp->t_pgrp = NULL;
685           }
686           lwkt_reltoken(&prg->proc_token);
687 
688           sess_rele(pgrp->pg_session);
689           kfree(pgrp, M_PGRP);
690 }
691 
692 /*
693  * Locate a process group by number.  The returned process group will be
694  * referenced w/pgref() and must be released with pgrel() (or assigned
695  * somewhere if you wish to keep the reference).
696  *
697  * No requirements.
698  */
699 struct pgrp *
pgfind(pid_t pgid)700 pgfind(pid_t pgid)
701 {
702           struct pgrp *pgrp;
703           procglob_t *prg;
704           int n;
705 
706           n = PGRP_HASH(pgid);
707           prg = &procglob[n];
708           lwkt_gettoken_shared(&prg->proc_token);
709 
710           LIST_FOREACH(pgrp, &prg->allpgrp, pg_list) {
711                     if (pgrp->pg_id == pgid) {
712                               refcount_acquire(&pgrp->pg_refs);
713                               lwkt_reltoken(&prg->proc_token);
714                               return (pgrp);
715                     }
716           }
717           lwkt_reltoken(&prg->proc_token);
718           return (NULL);
719 }
720 
721 /*
722  * Move p to a new or existing process group (and session)
723  *
724  * No requirements.
725  */
726 int
enterpgrp(struct proc * p,pid_t pgid,int mksess)727 enterpgrp(struct proc *p, pid_t pgid, int mksess)
728 {
729           struct pgrp *pgrp;
730           struct pgrp *opgrp;
731           int error;
732 
733           pgrp = pgfind(pgid);
734 
735           KASSERT(pgrp == NULL || !mksess,
736                     ("enterpgrp: setsid into non-empty pgrp"));
737           KASSERT(!SESS_LEADER(p),
738                     ("enterpgrp: session leader attempted setpgrp"));
739 
740           if (pgrp == NULL) {
741                     pid_t savepid = p->p_pid;
742                     struct proc *np;
743                     procglob_t *prg;
744                     int n;
745 
746                     /*
747                      * new process group
748                      */
749                     KASSERT(p->p_pid == pgid,
750                               ("enterpgrp: new pgrp and pid != pgid"));
751                     pgrp = kmalloc(sizeof(struct pgrp), M_PGRP, M_WAITOK | M_ZERO);
752                     pgrp->pg_id = pgid;
753                     LIST_INIT(&pgrp->pg_members);
754                     pgrp->pg_jobc = 0;
755                     SLIST_INIT(&pgrp->pg_sigiolst);
756                     lwkt_token_init(&pgrp->pg_token, "pgrp_token");
757                     refcount_init(&pgrp->pg_refs, 1);
758                     lockinit(&pgrp->pg_lock, "pgwt", 0, 0);
759 
760                     n = PGRP_HASH(pgid);
761                     prg = &procglob[n];
762 
763                     if ((np = pfindn(savepid)) == NULL || np != p) {
764                               lwkt_reltoken(&prg->proc_token);
765                               error = ESRCH;
766                               kfree(pgrp, M_PGRP);
767                               goto fatal;
768                     }
769 
770                     lwkt_gettoken(&prg->proc_token);
771                     if (mksess) {
772                               struct session *sess;
773 
774                               /*
775                                * new session
776                                */
777                               sess = kmalloc(sizeof(struct session), M_SESSION,
778                                                M_WAITOK | M_ZERO);
779                               lwkt_gettoken(&p->p_token);
780                               sess->s_prg = prg;
781                               sess->s_leader = p;
782                               sess->s_sid = p->p_pid;
783                               sess->s_count = 1;
784                               sess->s_ttyvp = NULL;
785                               sess->s_ttyp = NULL;
786                               bcopy(p->p_session->s_login, sess->s_login,
787                                     sizeof(sess->s_login));
788                               pgrp->pg_session = sess;
789                               KASSERT(p == curproc,
790                                         ("enterpgrp: mksession and p != curproc"));
791                               p->p_flags &= ~P_CONTROLT;
792                               LIST_INSERT_HEAD(&prg->allsess, sess, s_list);
793                               lwkt_reltoken(&p->p_token);
794                     } else {
795                               lwkt_gettoken(&p->p_token);
796                               pgrp->pg_session = p->p_session;
797                               sess_hold(pgrp->pg_session);
798                               lwkt_reltoken(&p->p_token);
799                     }
800                     LIST_INSERT_HEAD(&prg->allpgrp, pgrp, pg_list);
801 
802                     lwkt_reltoken(&prg->proc_token);
803           } else if (pgrp == p->p_pgrp) {
804                     pgrel(pgrp);
805                     goto done;
806           } /* else pgfind() referenced the pgrp */
807 
808           lwkt_gettoken(&pgrp->pg_token);
809           lwkt_gettoken(&p->p_token);
810 
811           /*
812            * Replace p->p_pgrp, handling any races that occur.
813            */
814           while ((opgrp = p->p_pgrp) != NULL) {
815                     pgref(opgrp);
816                     lwkt_gettoken(&opgrp->pg_token);
817                     if (opgrp != p->p_pgrp) {
818                               lwkt_reltoken(&opgrp->pg_token);
819                               pgrel(opgrp);
820                               continue;
821                     }
822                     LIST_REMOVE(p, p_pglist);
823                     break;
824           }
825           p->p_pgrp = pgrp;
826           LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist);
827 
828           /*
829            * Adjust eligibility of affected pgrps to participate in job control.
830            * Increment eligibility counts before decrementing, otherwise we
831            * could reach 0 spuriously during the first call.
832            */
833           fixjobc(p, pgrp, 1);
834           if (opgrp) {
835                     fixjobc(p, opgrp, 0);
836                     lwkt_reltoken(&opgrp->pg_token);
837                     pgrel(opgrp);       /* manual pgref */
838                     pgrel(opgrp);       /* p->p_pgrp ref */
839           }
840           lwkt_reltoken(&p->p_token);
841           lwkt_reltoken(&pgrp->pg_token);
842 done:
843           error = 0;
844 fatal:
845           return (error);
846 }
847 
848 /*
849  * Remove process from process group
850  *
851  * No requirements.
852  */
853 int
leavepgrp(struct proc * p)854 leavepgrp(struct proc *p)
855 {
856           struct pgrp *pg = p->p_pgrp;
857 
858           lwkt_gettoken(&p->p_token);
859           while ((pg = p->p_pgrp) != NULL) {
860                     pgref(pg);
861                     lwkt_gettoken(&pg->pg_token);
862                     if (p->p_pgrp != pg) {
863                               lwkt_reltoken(&pg->pg_token);
864                               pgrel(pg);
865                               continue;
866                     }
867                     p->p_pgrp = NULL;
868                     LIST_REMOVE(p, p_pglist);
869                     lwkt_reltoken(&pg->pg_token);
870                     pgrel(pg);          /* manual pgref */
871                     pgrel(pg);          /* p->p_pgrp ref */
872                     break;
873           }
874           lwkt_reltoken(&p->p_token);
875 
876           return (0);
877 }
878 
879 /*
880  * Adjust the ref count on a session structure.  When the ref count falls to
881  * zero the tty is disassociated from the session and the session structure
882  * is freed.  Note that tty assocation is not itself ref-counted.
883  *
884  * No requirements.
885  */
886 void
sess_hold(struct session * sp)887 sess_hold(struct session *sp)
888 {
889           atomic_add_int(&sp->s_count, 1);
890 }
891 
892 /*
893  * No requirements.
894  */
895 void
sess_rele(struct session * sess)896 sess_rele(struct session *sess)
897 {
898           procglob_t *prg;
899           struct tty *tp;
900           int count;
901           int n;
902 
903           n = SESS_HASH(sess->s_sid);
904           prg = &procglob[n];
905 
906           for (;;) {
907                     count = sess->s_count;
908                     cpu_ccfence();
909                     KKASSERT(count > 0);
910                     if (count == 1) {
911                               lwkt_gettoken(&prg->proc_token);
912                               if (atomic_cmpset_int(&sess->s_count, 1, 0))
913                                         break;
914                               lwkt_reltoken(&prg->proc_token);
915                               /* retry */
916                     } else {
917                               if (atomic_cmpset_int(&sess->s_count, count, count - 1))
918                                         return;
919                               /* retry */
920                     }
921           }
922 
923           /*
924            * Successful 1->0 transition and prg->proc_token is held.
925            */
926           LIST_REMOVE(sess, s_list);
927           if (pid_doms[sess->s_sid % PIDSEL_DOMAINS] != (uint8_t)time_second)
928                     pid_doms[sess->s_sid % PIDSEL_DOMAINS] = (uint8_t)time_second;
929 
930           if (sess->s_ttyp && sess->s_ttyp->t_session) {
931 #ifdef TTY_DO_FULL_CLOSE
932                     /* FULL CLOSE, see ttyclearsession() */
933                     KKASSERT(sess->s_ttyp->t_session == sess);
934                     sess->s_ttyp->t_session = NULL;
935 #else
936                     /* HALF CLOSE, see ttyclearsession() */
937                     if (sess->s_ttyp->t_session == sess)
938                               sess->s_ttyp->t_session = NULL;
939 #endif
940           }
941           if ((tp = sess->s_ttyp) != NULL) {
942                     sess->s_ttyp = NULL;
943                     ttyunhold(tp);
944           }
945           lwkt_reltoken(&prg->proc_token);
946 
947           kfree(sess, M_SESSION);
948 }
949 
950 /*
951  * Adjust pgrp jobc counters when specified process changes process group.
952  * We count the number of processes in each process group that "qualify"
953  * the group for terminal job control (those with a parent in a different
954  * process group of the same session).  If that count reaches zero, the
955  * process group becomes orphaned.  Check both the specified process'
956  * process group and that of its children.
957  * entering == 0 => p is leaving specified group.
958  * entering == 1 => p is entering specified group.
959  *
960  * No requirements.
961  */
962 void
fixjobc(struct proc * p,struct pgrp * pgrp,int entering)963 fixjobc(struct proc *p, struct pgrp *pgrp, int entering)
964 {
965           struct pgrp *hispgrp;
966           struct session *mysession;
967           struct proc *np;
968 
969           /*
970            * Check p's parent to see whether p qualifies its own process
971            * group; if so, adjust count for p's process group.
972            */
973           lwkt_gettoken(&p->p_token);   /* p_children scan */
974           lwkt_gettoken(&pgrp->pg_token);
975 
976           mysession = pgrp->pg_session;
977           if ((hispgrp = p->p_pptr->p_pgrp) != pgrp &&
978               hispgrp->pg_session == mysession) {
979                     if (entering)
980                               pgrp->pg_jobc++;
981                     else if (--pgrp->pg_jobc == 0)
982                               orphanpg(pgrp);
983           }
984 
985           /*
986            * Check this process' children to see whether they qualify
987            * their process groups; if so, adjust counts for children's
988            * process groups.
989            */
990           LIST_FOREACH(np, &p->p_children, p_sibling) {
991                     PHOLD(np);
992                     lwkt_gettoken(&np->p_token);
993                     if ((hispgrp = np->p_pgrp) != pgrp &&
994                         hispgrp->pg_session == mysession &&
995                         np->p_stat != SZOMB) {
996                               pgref(hispgrp);
997                               lwkt_gettoken(&hispgrp->pg_token);
998                               if (entering)
999                                         hispgrp->pg_jobc++;
1000                               else if (--hispgrp->pg_jobc == 0)
1001                                         orphanpg(hispgrp);
1002                               lwkt_reltoken(&hispgrp->pg_token);
1003                               pgrel(hispgrp);
1004                     }
1005                     lwkt_reltoken(&np->p_token);
1006                     PRELE(np);
1007           }
1008           KKASSERT(pgrp->pg_refs > 0);
1009           lwkt_reltoken(&pgrp->pg_token);
1010           lwkt_reltoken(&p->p_token);
1011 }
1012 
1013 /*
1014  * A process group has become orphaned;
1015  * if there are any stopped processes in the group,
1016  * hang-up all process in that group.
1017  *
1018  * The caller must hold pg_token.
1019  */
1020 static void
orphanpg(struct pgrp * pg)1021 orphanpg(struct pgrp *pg)
1022 {
1023           struct proc *p;
1024 
1025           LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1026                     if (p->p_stat == SSTOP) {
1027                               LIST_FOREACH(p, &pg->pg_members, p_pglist) {
1028                                         ksignal(p, SIGHUP);
1029                                         ksignal(p, SIGCONT);
1030                               }
1031                               return;
1032                     }
1033           }
1034 }
1035 
1036 /*
1037  * Add a new process to the allproc list and the PID hash.  This
1038  * also assigns a pid to the new process.
1039  *
1040  * No requirements.
1041  */
1042 void
proc_add_allproc(struct proc * p)1043 proc_add_allproc(struct proc *p)
1044 {
1045           int random_offset;
1046 
1047           if ((random_offset = randompid) != 0) {
1048                     read_random(&random_offset, sizeof(random_offset), 1);
1049                     random_offset = (random_offset & 0x7FFFFFFF) % randompid;
1050           }
1051           proc_makepid(p, random_offset);
1052 }
1053 
1054 /*
1055  * Calculate a new process pid.  This function is integrated into
1056  * proc_add_allproc() to guarentee that the new pid is not reused before
1057  * the new process can be added to the allproc list.
1058  *
1059  * p_pid is assigned and the process is added to the allproc hash table
1060  *
1061  * WARNING! We need to allocate PIDs sequentially during early boot.
1062  *            In particular, init needs to have a pid of 1.
1063  */
1064 static
1065 void
proc_makepid(struct proc * p,int random_offset)1066 proc_makepid(struct proc *p, int random_offset)
1067 {
1068           static pid_t nextpid = 1;     /* heuristic, allowed to race */
1069           procglob_t *prg;
1070           struct pgrp *pg;
1071           struct proc *ps;
1072           struct session *sess;
1073           pid_t base;
1074           int8_t delta8;
1075           int retries;
1076           int n;
1077 
1078           /*
1079            * Select the next pid base candidate.
1080            *
1081            * Check cyclement, do not allow a pid < 100.
1082            */
1083           retries = 0;
1084 retry:
1085           base = atomic_fetchadd_int(&nextpid, 1) + random_offset;
1086           if (base <= 0 || base >= PID_MAX) {
1087                     base = base % PID_MAX;
1088                     if (base < 0)
1089                               base = 100;
1090                     if (base < 100)
1091                               base += 100;
1092                     nextpid = base;               /* reset (SMP race ok) */
1093           }
1094 
1095           /*
1096            * Do not allow a base pid to be selected from a domain that has
1097            * recently seen a pid/pgid/sessid reap.  Sleep a little if we looped
1098            * through all available domains.
1099            *
1100            * WARNING: We want the early pids to be allocated linearly,
1101            *            particularly pid 1 and pid 2.
1102            */
1103           if (++retries >= PIDSEL_DOMAINS)
1104                     tsleep(&nextpid, 0, "makepid", 1);
1105           if (base >= 100) {
1106                     delta8 = (int8_t)time_second -
1107                                (int8_t)pid_doms[base % PIDSEL_DOMAINS];
1108                     if (delta8 >= 0 && delta8 <= PIDDOM_DELAY) {
1109                               ++pid_domain_skips;
1110                               goto retry;
1111                     }
1112           }
1113 
1114           /*
1115            * Calculate a hash index and find an unused process id within
1116            * the table, looping if we cannot find one.
1117            *
1118            * The inner loop increments by ALLPROC_HSIZE which keeps the
1119            * PID at the same pid_doms[] index as well as the same hash index.
1120            */
1121           n = ALLPROC_HASH(base);
1122           prg = &procglob[n];
1123           lwkt_gettoken(&prg->proc_token);
1124 
1125 restart1:
1126           LIST_FOREACH(ps, &prg->allproc, p_list) {
1127                     if (ps->p_pid == base) {
1128                               base += ALLPROC_HSIZE;
1129                               if (base >= PID_MAX) {
1130                                         lwkt_reltoken(&prg->proc_token);
1131                                         goto retry;
1132                               }
1133                               ++pid_inner_skips;
1134                               goto restart1;
1135                     }
1136           }
1137           LIST_FOREACH(pg, &prg->allpgrp, pg_list) {
1138                     if (pg->pg_id == base) {
1139                               base += ALLPROC_HSIZE;
1140                               if (base >= PID_MAX) {
1141                                         lwkt_reltoken(&prg->proc_token);
1142                                         goto retry;
1143                               }
1144                               ++pid_inner_skips;
1145                               goto restart1;
1146                     }
1147           }
1148           LIST_FOREACH(sess, &prg->allsess, s_list) {
1149                     if (sess->s_sid == base) {
1150                               base += ALLPROC_HSIZE;
1151                               if (base >= PID_MAX) {
1152                                         lwkt_reltoken(&prg->proc_token);
1153                                         goto retry;
1154                               }
1155                               ++pid_inner_skips;
1156                               goto restart1;
1157                     }
1158           }
1159 
1160           /*
1161            * Assign the pid and insert the process.
1162            */
1163           p->p_pid = base;
1164           LIST_INSERT_HEAD(&prg->allproc, p, p_list);
1165           lwkt_reltoken(&prg->proc_token);
1166 }
1167 
1168 /*
1169  * Called from exit1 to place the process into a zombie state.
1170  * The process is removed from the pid hash and p_stat is set
1171  * to SZOMB.  Normal pfind[n]() calls will not find it any more.
1172  *
1173  * Caller must hold p->p_token.  We are required to wait until p_lock
1174  * becomes zero before we can manipulate the list, allowing allproc
1175  * scans to guarantee consistency during a list scan.
1176  */
1177 void
proc_move_allproc_zombie(struct proc * p)1178 proc_move_allproc_zombie(struct proc *p)
1179 {
1180           procglob_t *prg;
1181           int n;
1182 
1183           n = ALLPROC_HASH(p->p_pid);
1184           prg = &procglob[n];
1185           PSTALL(p, "reap1", 0);
1186           lwkt_gettoken(&prg->proc_token);
1187 
1188           PSTALL(p, "reap1a", 0);
1189           p->p_stat = SZOMB;
1190 
1191           lwkt_reltoken(&prg->proc_token);
1192           dsched_exit_proc(p);
1193 }
1194 
1195 /*
1196  * This routine is called from kern_wait() and will remove the process
1197  * from the zombie list and the sibling list.  This routine will block
1198  * if someone has a lock on the proces (p_lock).
1199  *
1200  * Caller must hold p->p_token.  We are required to wait until p_lock
1201  * becomes one before we can manipulate the list, allowing allproc
1202  * scans to guarantee consistency during a list scan.
1203  *
1204  * Assumes caller has one ref.
1205  */
1206 void
proc_remove_zombie(struct proc * p)1207 proc_remove_zombie(struct proc *p)
1208 {
1209           procglob_t *prg;
1210           int n;
1211 
1212           n = ALLPROC_HASH(p->p_pid);
1213           prg = &procglob[n];
1214 
1215           PSTALL(p, "reap2", 1);
1216           lwkt_gettoken(&prg->proc_token);
1217           PSTALL(p, "reap2a", 1);
1218           LIST_REMOVE(p, p_list);                 /* from remove master list */
1219           LIST_REMOVE(p, p_sibling);    /* and from sibling list */
1220           p->p_pptr = NULL;
1221           p->p_ppid = 0;
1222           if (pid_doms[p->p_pid % PIDSEL_DOMAINS] != (uint8_t)time_second)
1223                     pid_doms[p->p_pid % PIDSEL_DOMAINS] = (uint8_t)time_second;
1224           lwkt_reltoken(&prg->proc_token);
1225 }
1226 
1227 /*
1228  * Handle various requirements prior to returning to usermode.  Called from
1229  * platform trap and system call code.
1230  */
1231 void
lwpuserret(struct lwp * lp)1232 lwpuserret(struct lwp *lp)
1233 {
1234           struct proc *p = lp->lwp_proc;
1235 
1236           if (lp->lwp_mpflags & LWP_MP_VNLRU) {
1237                     atomic_clear_int(&lp->lwp_mpflags, LWP_MP_VNLRU);
1238                     allocvnode_gc();
1239           }
1240           if (lp->lwp_mpflags & LWP_MP_WEXIT) {
1241                     lwkt_gettoken(&p->p_token);
1242                     lwp_exit(0, NULL);
1243                     lwkt_reltoken(&p->p_token);     /* NOT REACHED */
1244           }
1245 }
1246 
1247 /*
1248  * Kernel threads run from user processes can also accumulate deferred
1249  * actions which need to be acted upon.  Callers include:
1250  *
1251  * nfsd             - Can allocate lots of vnodes
1252  */
1253 void
lwpkthreaddeferred(void)1254 lwpkthreaddeferred(void)
1255 {
1256           struct lwp *lp = curthread->td_lwp;
1257 
1258           if (lp) {
1259                     if (lp->lwp_mpflags & LWP_MP_VNLRU) {
1260                               atomic_clear_int(&lp->lwp_mpflags, LWP_MP_VNLRU);
1261                               allocvnode_gc();
1262                     }
1263           }
1264 }
1265 
1266 void
proc_usermap(struct proc * p,int invfork)1267 proc_usermap(struct proc *p, int invfork)
1268 {
1269           struct sys_upmap *upmap;
1270 
1271           lwkt_gettoken(&p->p_token);
1272           upmap = kmalloc(roundup2(sizeof(*upmap), PAGE_SIZE), M_UPMAP,
1273                               M_WAITOK | M_ZERO);
1274           if (p->p_upmap == NULL && (p->p_flags & P_POSTEXIT) == 0) {
1275                     upmap->header[0].type = UKPTYPE_VERSION;
1276                     upmap->header[0].offset = offsetof(struct sys_upmap, version);
1277                     upmap->header[1].type = UPTYPE_RUNTICKS;
1278                     upmap->header[1].offset = offsetof(struct sys_upmap, runticks);
1279                     upmap->header[2].type = UPTYPE_FORKID;
1280                     upmap->header[2].offset = offsetof(struct sys_upmap, forkid);
1281                     upmap->header[3].type = UPTYPE_PID;
1282                     upmap->header[3].offset = offsetof(struct sys_upmap, pid);
1283                     upmap->header[4].type = UPTYPE_PROC_TITLE;
1284                     upmap->header[4].offset = offsetof(struct sys_upmap,proc_title);
1285                     upmap->header[5].type = UPTYPE_INVFORK;
1286                     upmap->header[5].offset = offsetof(struct sys_upmap, invfork);
1287 
1288                     upmap->version = UPMAP_VERSION;
1289                     upmap->pid = p->p_pid;
1290                     upmap->forkid = p->p_forkid;
1291                     upmap->invfork = invfork;
1292                     p->p_upmap = upmap;
1293           } else {
1294                     kfree(upmap, M_UPMAP);
1295           }
1296           lwkt_reltoken(&p->p_token);
1297 }
1298 
1299 void
proc_userunmap(struct proc * p)1300 proc_userunmap(struct proc *p)
1301 {
1302           struct sys_upmap *upmap;
1303 
1304           lwkt_gettoken(&p->p_token);
1305           if ((upmap = p->p_upmap) != NULL) {
1306                     p->p_upmap = NULL;
1307                     kfree(upmap, M_UPMAP);
1308           }
1309           lwkt_reltoken(&p->p_token);
1310 }
1311 
1312 /*
1313  * Called when the per-thread user/kernel shared page needs to be
1314  * allocated.  The function refuses to allocate the page if the
1315  * thread is exiting to avoid races against lwp_userunmap().
1316  */
1317 void
lwp_usermap(struct lwp * lp,int invfork)1318 lwp_usermap(struct lwp *lp, int invfork)
1319 {
1320           struct sys_lpmap *lpmap;
1321 
1322           lwkt_gettoken(&lp->lwp_token);
1323 
1324           lpmap = kmalloc(roundup2(sizeof(*lpmap), PAGE_SIZE), M_UPMAP,
1325                               M_WAITOK | M_ZERO);
1326           if (lp->lwp_lpmap == NULL && (lp->lwp_mpflags & LWP_MP_WEXIT) == 0) {
1327                     lpmap->header[0].type = UKPTYPE_VERSION;
1328                     lpmap->header[0].offset = offsetof(struct sys_lpmap, version);
1329                     lpmap->header[1].type = LPTYPE_BLOCKALLSIGS;
1330                     lpmap->header[1].offset = offsetof(struct sys_lpmap,
1331                                                                blockallsigs);
1332                     lpmap->header[2].type = LPTYPE_THREAD_TITLE;
1333                     lpmap->header[2].offset = offsetof(struct sys_lpmap,
1334                                                                thread_title);
1335                     lpmap->header[3].type = LPTYPE_THREAD_TID;
1336                     lpmap->header[3].offset = offsetof(struct sys_lpmap, tid);
1337 
1338                     lpmap->version = LPMAP_VERSION;
1339                     lpmap->tid = lp->lwp_tid;
1340                     lp->lwp_lpmap = lpmap;
1341           } else {
1342                     kfree(lpmap, M_UPMAP);
1343           }
1344           lwkt_reltoken(&lp->lwp_token);
1345 }
1346 
1347 /*
1348  * Called when a LWP (but not necessarily the whole process) exits.
1349  * Called when a process execs (after all other threads have been killed).
1350  *
1351  * lwp-specific mappings must be removed.  If userland didn't do it, then
1352  * we have to.  Otherwise we could end-up disclosing kernel memory due to
1353  * the ad-hoc pmap mapping.
1354  */
1355 void
lwp_userunmap(struct lwp * lp)1356 lwp_userunmap(struct lwp *lp)
1357 {
1358           struct sys_lpmap *lpmap;
1359           struct vm_map *map;
1360           struct vm_map_backing *ba;
1361           struct vm_map_backing copy;
1362 
1363           lwkt_gettoken(&lp->lwp_token);
1364           map = &lp->lwp_proc->p_vmspace->vm_map;
1365           lpmap = lp->lwp_lpmap;
1366           lp->lwp_lpmap = NULL;
1367 
1368           spin_lock(&lp->lwp_spin);
1369           while ((ba = TAILQ_FIRST(&lp->lwp_lpmap_backing_list)) != NULL) {
1370                     copy = *ba;
1371                     spin_unlock(&lp->lwp_spin);
1372 
1373                     lwkt_gettoken(&map->token);
1374                     vm_map_remove(map, copy.start, copy.end);
1375                     lwkt_reltoken(&map->token);
1376 
1377                     spin_lock(&lp->lwp_spin);
1378           }
1379           spin_unlock(&lp->lwp_spin);
1380 
1381           if (lpmap)
1382                     kfree(lpmap, M_UPMAP);
1383           lwkt_reltoken(&lp->lwp_token);
1384 }
1385 
1386 /*
1387  * Scan all processes on the allproc list.  The process is automatically
1388  * held for the callback.  A return value of -1 terminates the loop.
1389  * Zombie procs are skipped.
1390  *
1391  * The callback is made with the process held and proc_token held.
1392  *
1393  * We limit the scan to the number of processes as-of the start of
1394  * the scan so as not to get caught up in an endless loop if new processes
1395  * are created more quickly than we can scan the old ones.  Add a little
1396  * slop to try to catch edge cases since nprocs can race.
1397  *
1398  * No requirements.
1399  */
1400 void
allproc_scan(int (* callback)(struct proc *,void *),void * data,int segmented)1401 allproc_scan(int (*callback)(struct proc *, void *), void *data, int segmented)
1402 {
1403           int limit = nprocs + ncpus;
1404           struct proc *p;
1405           int ns;
1406           int ne;
1407           int r;
1408           int n;
1409 
1410           if (segmented) {
1411                     int id = mycpu->gd_cpuid;
1412                     ns = id * ALLPROC_HSIZE / ncpus;
1413                     ne = (id + 1) * ALLPROC_HSIZE / ncpus;
1414           } else {
1415                     ns = 0;
1416                     ne = ALLPROC_HSIZE;
1417           }
1418 
1419           /*
1420            * prg->proc_token protects the allproc list and PHOLD() prevents the
1421            * process from being removed from the allproc list or the zombproc
1422            * list.
1423            */
1424           for (n = ns; n < ne; ++n) {
1425                     procglob_t *prg = &procglob[n];
1426                     if (LIST_FIRST(&prg->allproc) == NULL)
1427                               continue;
1428                     lwkt_gettoken(&prg->proc_token);
1429                     LIST_FOREACH(p, &prg->allproc, p_list) {
1430                               if (p->p_stat == SZOMB)
1431                                         continue;
1432                               PHOLD(p);
1433                               r = callback(p, data);
1434                               PRELE(p);
1435                               if (r < 0)
1436                                         break;
1437                               if (--limit < 0)
1438                                         break;
1439                     }
1440                     lwkt_reltoken(&prg->proc_token);
1441 
1442                     /*
1443                      * Check if asked to stop early
1444                      */
1445                     if (p)
1446                               break;
1447           }
1448 }
1449 
1450 /*
1451  * Scan all lwps of processes on the allproc list.  The lwp is automatically
1452  * held for the callback.  A return value of -1 terminates the loop.
1453  *
1454  * The callback is made with the proces and lwp both held, and proc_token held.
1455  *
1456  * No requirements.
1457  */
1458 void
alllwp_scan(int (* callback)(struct lwp *,void *),void * data,int segmented)1459 alllwp_scan(int (*callback)(struct lwp *, void *), void *data, int segmented)
1460 {
1461           struct proc *p;
1462           struct lwp *lp;
1463           int ns;
1464           int ne;
1465           int r = 0;
1466           int n;
1467 
1468           if (segmented) {
1469                     int id = mycpu->gd_cpuid;
1470                     ns = id * ALLPROC_HSIZE / ncpus;
1471                     ne = (id + 1) * ALLPROC_HSIZE / ncpus;
1472           } else {
1473                     ns = 0;
1474                     ne = ALLPROC_HSIZE;
1475           }
1476 
1477           for (n = ns; n < ne; ++n) {
1478                     procglob_t *prg = &procglob[n];
1479 
1480                     if (LIST_FIRST(&prg->allproc) == NULL)
1481                               continue;
1482                     lwkt_gettoken(&prg->proc_token);
1483                     LIST_FOREACH(p, &prg->allproc, p_list) {
1484                               if (p->p_stat == SZOMB)
1485                                         continue;
1486                               PHOLD(p);
1487                               lwkt_gettoken(&p->p_token);
1488                               FOREACH_LWP_IN_PROC(lp, p) {
1489                                         LWPHOLD(lp);
1490                                         r = callback(lp, data);
1491                                         LWPRELE(lp);
1492                               }
1493                               lwkt_reltoken(&p->p_token);
1494                               PRELE(p);
1495                               if (r < 0)
1496                                         break;
1497                     }
1498                     lwkt_reltoken(&prg->proc_token);
1499 
1500                     /*
1501                      * Asked to exit early
1502                      */
1503                     if (p)
1504                               break;
1505           }
1506 }
1507 
1508 /*
1509  * Scan all processes on the zombproc list.  The process is automatically
1510  * held for the callback.  A return value of -1 terminates the loop.
1511  *
1512  * No requirements.
1513  * The callback is made with the proces held and proc_token held.
1514  */
1515 void
zombproc_scan(int (* callback)(struct proc *,void *),void * data)1516 zombproc_scan(int (*callback)(struct proc *, void *), void *data)
1517 {
1518           struct proc *p;
1519           int r;
1520           int n;
1521 
1522           /*
1523            * prg->proc_token protects the allproc list and PHOLD() prevents the
1524            * process from being removed from the allproc list or the zombproc
1525            * list.
1526            */
1527           for (n = 0; n < ALLPROC_HSIZE; ++n) {
1528                     procglob_t *prg = &procglob[n];
1529 
1530                     if (LIST_FIRST(&prg->allproc) == NULL)
1531                               continue;
1532                     lwkt_gettoken(&prg->proc_token);
1533                     LIST_FOREACH(p, &prg->allproc, p_list) {
1534                               if (p->p_stat != SZOMB)
1535                                         continue;
1536                               PHOLD(p);
1537                               r = callback(p, data);
1538                               PRELE(p);
1539                               if (r < 0)
1540                                         break;
1541                     }
1542                     lwkt_reltoken(&prg->proc_token);
1543 
1544                     /*
1545                      * Check if asked to stop early
1546                      */
1547                     if (p)
1548                               break;
1549           }
1550 }
1551 
1552 #include "opt_ddb.h"
1553 #ifdef DDB
1554 #include <ddb/ddb.h>
1555 
1556 /*
1557  * Debugging only
1558  */
DB_SHOW_COMMAND(pgrpdump,pgrpdump)1559 DB_SHOW_COMMAND(pgrpdump, pgrpdump)
1560 {
1561           struct pgrp *pgrp;
1562           struct proc *p;
1563           procglob_t *prg;
1564           int i;
1565 
1566           for (i = 0; i < ALLPROC_HSIZE; ++i) {
1567                     prg = &procglob[i];
1568 
1569                     if (LIST_EMPTY(&prg->allpgrp))
1570                               continue;
1571                     kprintf("\tindx %d\n", i);
1572                     LIST_FOREACH(pgrp, &prg->allpgrp, pg_list) {
1573                               kprintf("\tpgrp %p, pgid %ld, sess %p, "
1574                                         "sesscnt %d, mem %p\n",
1575                                         (void *)pgrp, (long)pgrp->pg_id,
1576                                         (void *)pgrp->pg_session,
1577                                         pgrp->pg_session->s_count,
1578                                         (void *)LIST_FIRST(&pgrp->pg_members));
1579                               LIST_FOREACH(p, &pgrp->pg_members, p_pglist) {
1580                                         kprintf("\t\tpid %ld addr %p pgrp %p\n",
1581                                                   (long)p->p_pid, (void *)p,
1582                                                   (void *)p->p_pgrp);
1583                               }
1584                     }
1585           }
1586 }
1587 #endif /* DDB */
1588 
1589 /*
1590  * The caller must hold proc_token.
1591  */
1592 static int
sysctl_out_proc(struct proc * p,struct sysctl_req * req,int flags)1593 sysctl_out_proc(struct proc *p, struct sysctl_req *req, int flags)
1594 {
1595           struct kinfo_proc ki;
1596           struct lwp *lp;
1597           int skip_lwp = 0;
1598           int had_output = 0;
1599           int error;
1600 
1601           bzero(&ki, sizeof(ki));
1602           lwkt_gettoken_shared(&p->p_token);
1603           fill_kinfo_proc(p, &ki);
1604           if ((flags & KERN_PROC_FLAG_LWP) == 0)
1605                     skip_lwp = 1;
1606           error = 0;
1607           FOREACH_LWP_IN_PROC(lp, p) {
1608                     LWPHOLD(lp);
1609                     fill_kinfo_lwp(lp, &ki.kp_lwp);
1610                     had_output = 1;
1611                     if (skip_lwp == 0) {
1612                               error = SYSCTL_OUT(req, &ki, sizeof(ki));
1613                               bzero(&ki.kp_lwp, sizeof(ki.kp_lwp));
1614                     }
1615                     LWPRELE(lp);
1616                     if (error)
1617                               break;
1618           }
1619           lwkt_reltoken(&p->p_token);
1620 
1621           /*
1622            * If aggregating threads, set the tid field to -1.
1623            */
1624           if (skip_lwp)
1625                     ki.kp_lwp.kl_tid = -1;
1626 
1627           /*
1628            * We need to output at least the proc, even if there is no lwp.
1629            * If skip_lwp is non-zero we aggregated the lwps and need to output
1630            * the result.
1631            */
1632           if (had_output == 0 || skip_lwp) {
1633                     error = SYSCTL_OUT(req, &ki, sizeof(ki));
1634           }
1635           return (error);
1636 }
1637 
1638 /*
1639  * The caller must hold proc_token.
1640  */
1641 static int
sysctl_out_proc_kthread(struct thread * td,struct sysctl_req * req)1642 sysctl_out_proc_kthread(struct thread *td, struct sysctl_req *req)
1643 {
1644           struct kinfo_proc ki;
1645           int error;
1646 
1647           fill_kinfo_proc_kthread(td, &ki);
1648           error = SYSCTL_OUT(req, &ki, sizeof(ki));
1649 
1650           return (error);
1651 }
1652 
1653 /*
1654  * No requirements.
1655  */
1656 static int
sysctl_kern_proc(SYSCTL_HANDLER_ARGS)1657 sysctl_kern_proc(SYSCTL_HANDLER_ARGS)
1658 {
1659           int *name = (int *)arg1;
1660           int oid = oidp->oid_number;
1661           u_int namelen = arg2;
1662           struct proc *p;
1663           struct thread *td;
1664           struct thread *marker;
1665           int flags = 0;
1666           int error = 0;
1667           int n;
1668           int origcpu;
1669           struct ucred *cr1 = curproc->p_ucred;
1670           struct ucred *crcache = NULL;
1671 
1672           flags = oid & KERN_PROC_FLAGMASK;
1673           oid &= ~KERN_PROC_FLAGMASK;
1674 
1675           if ((oid == KERN_PROC_ALL && namelen != 0) ||
1676               (oid != KERN_PROC_ALL && namelen != 1))
1677           {
1678                     return (EINVAL);
1679           }
1680 
1681           /*
1682            * proc_token protects the allproc list and PHOLD() prevents the
1683            * process from being removed from the allproc list or the zombproc
1684            * list.
1685            */
1686           if (oid == KERN_PROC_PID) {
1687                     p = pfind((pid_t)name[0]);
1688                     if (p) {
1689                               crcache = pcredcache(crcache, p);
1690                               if (PRISON_CHECK(cr1, crcache))
1691                                         error = sysctl_out_proc(p, req, flags);
1692                               PRELE(p);
1693                     }
1694                     goto post_threads;
1695           }
1696           p = NULL;
1697 
1698           if (!req->oldptr) {
1699                     /* overestimate by 5 procs */
1700                     error = SYSCTL_OUT(req, 0, sizeof (struct kinfo_proc) * 5);
1701                     if (error)
1702                               goto post_threads;
1703           }
1704 
1705           for (n = 0; n < ALLPROC_HSIZE; ++n) {
1706                     procglob_t *prg = &procglob[n];
1707 
1708                     if (LIST_EMPTY(&prg->allproc))
1709                               continue;
1710                     lwkt_gettoken_shared(&prg->proc_token);
1711                     LIST_FOREACH(p, &prg->allproc, p_list) {
1712                               /*
1713                                * Show a user only their processes.
1714                                */
1715                               if (ps_showallprocs == 0) {
1716                                         crcache = pcredcache(crcache, p);
1717                                         if (crcache == NULL ||
1718                                             p_trespass(cr1, crcache)) {
1719                                                   continue;
1720                                         }
1721                               }
1722 
1723                               /*
1724                                * Skip embryonic processes.
1725                                */
1726                               if (p->p_stat == SIDL)
1727                                         continue;
1728 
1729                               /*
1730                                * TODO - make more efficient (see notes below).
1731                                * do by session.
1732                                */
1733                               switch (oid) {
1734                               case KERN_PROC_PGRP:
1735                                         /* could do this by traversing pgrp */
1736                                         if (p->p_pgrp == NULL ||
1737                                             p->p_pgrp->pg_id != (pid_t)name[0])
1738                                                   continue;
1739                                         break;
1740 
1741                               case KERN_PROC_TTY:
1742                                         if ((p->p_flags & P_CONTROLT) == 0 ||
1743                                             p->p_session == NULL ||
1744                                             p->p_session->s_ttyp == NULL ||
1745                                             devid_from_dev(p->p_session->s_ttyp->t_dev) !=
1746                                                   (dev_t)name[0])
1747                                                   continue;
1748                                         break;
1749 
1750                               case KERN_PROC_UID:
1751                                         crcache = pcredcache(crcache, p);
1752                                         if (crcache == NULL ||
1753                                             crcache->cr_uid != (uid_t)name[0]) {
1754                                                   continue;
1755                                         }
1756                                         break;
1757 
1758                               case KERN_PROC_RUID:
1759                                         crcache = pcredcache(crcache, p);
1760                                         if (crcache == NULL ||
1761                                             crcache->cr_ruid != (uid_t)name[0]) {
1762                                                   continue;
1763                                         }
1764                                         break;
1765                               }
1766 
1767                               crcache = pcredcache(crcache, p);
1768                               if (!PRISON_CHECK(cr1, crcache))
1769                                         continue;
1770                               PHOLD(p);
1771                               error = sysctl_out_proc(p, req, flags);
1772                               PRELE(p);
1773                               if (error) {
1774                                         lwkt_reltoken(&prg->proc_token);
1775                                         goto post_threads;
1776                               }
1777                     }
1778                     lwkt_reltoken(&prg->proc_token);
1779           }
1780 
1781           /*
1782            * Iterate over all active cpus and scan their thread list.  Start
1783            * with the next logical cpu and end with our original cpu.  We
1784            * migrate our own thread to each target cpu in order to safely scan
1785            * its thread list.  In the last loop we migrate back to our original
1786            * cpu.
1787            */
1788           origcpu = mycpu->gd_cpuid;
1789           if (!ps_showallthreads || jailed(cr1))
1790                     goto post_threads;
1791 
1792           marker = kmalloc(sizeof(struct thread), M_TEMP, M_WAITOK|M_ZERO);
1793           marker->td_flags = TDF_MARKER;
1794           error = 0;
1795 
1796           for (n = 1; (flags & KERN_PROC_FLAG_LWKT) && n <= ncpus; ++n) {
1797                     globaldata_t rgd;
1798                     int nid;
1799 
1800                     nid = (origcpu + n) % ncpus;
1801                     if (CPUMASK_TESTBIT(smp_active_mask, nid) == 0)
1802                               continue;
1803                     rgd = globaldata_find(nid);
1804                     lwkt_setcpu_self(rgd);
1805 
1806                     crit_enter();
1807                     TAILQ_INSERT_TAIL(&rgd->gd_tdallq, marker, td_allq);
1808 
1809                     while ((td = TAILQ_PREV(marker, lwkt_queue, td_allq)) != NULL) {
1810                               TAILQ_REMOVE(&rgd->gd_tdallq, marker, td_allq);
1811                               TAILQ_INSERT_BEFORE(td, marker, td_allq);
1812                               if (td->td_flags & TDF_MARKER)
1813                                         continue;
1814                               if (td->td_proc)
1815                                         continue;
1816 
1817                               lwkt_hold(td);
1818                               crit_exit();
1819 
1820                               switch (oid) {
1821                               case KERN_PROC_PGRP:
1822                               case KERN_PROC_TTY:
1823                               case KERN_PROC_UID:
1824                               case KERN_PROC_RUID:
1825                                         break;
1826                               default:
1827                                         error = sysctl_out_proc_kthread(td, req);
1828                                         break;
1829                               }
1830                               lwkt_rele(td);
1831                               crit_enter();
1832                               if (error)
1833                                         break;
1834                     }
1835                     TAILQ_REMOVE(&rgd->gd_tdallq, marker, td_allq);
1836                     crit_exit();
1837 
1838                     if (error)
1839                               break;
1840           }
1841 
1842           /*
1843            * Userland scheduler expects us to return on the same cpu we
1844            * started on.
1845            */
1846           if (mycpu->gd_cpuid != origcpu)
1847                     lwkt_setcpu_self(globaldata_find(origcpu));
1848 
1849           kfree(marker, M_TEMP);
1850 
1851 post_threads:
1852           if (crcache)
1853                     crfree(crcache);
1854           return (error);
1855 }
1856 
1857 /*
1858  * This sysctl allows a process to retrieve the argument list or process
1859  * title for another process without groping around in the address space
1860  * of the other process.  It also allow a process to set its own "process
1861  * title to a string of its own choice.
1862  *
1863  * No requirements.
1864  */
1865 static int
sysctl_kern_proc_args(SYSCTL_HANDLER_ARGS)1866 sysctl_kern_proc_args(SYSCTL_HANDLER_ARGS)
1867 {
1868           int *name = (int*) arg1;
1869           u_int namelen = arg2;
1870           size_t n;
1871           struct proc *p;
1872           struct lwp *lp;
1873 #if 0
1874           struct pargs *opa;
1875 #endif
1876           struct pargs *pa;
1877           int error = 0;
1878           struct ucred *cr1 = curproc->p_ucred;
1879 
1880           if (namelen != 1 && namelen != 2)
1881                     return (EINVAL);
1882 
1883           lp = NULL;
1884           p = pfind((pid_t)name[0]);
1885           if (p == NULL)
1886                     goto done;
1887           lwkt_gettoken(&p->p_token);
1888 
1889           if (namelen == 2) {
1890                     lp = lwpfind(p, (lwpid_t)name[1]);
1891                     if (lp)
1892                               lwkt_gettoken(&lp->lwp_token);
1893           } else {
1894                     lp = NULL;
1895           }
1896 
1897           if ((!ps_argsopen) && p_trespass(cr1, p->p_ucred))
1898                     goto done;
1899 
1900           if (req->newptr && curproc != p) {
1901                     error = EPERM;
1902                     goto done;
1903           }
1904           if (req->oldptr) {
1905                     if (lp && lp->lwp_lpmap != NULL &&
1906                         lp->lwp_lpmap->thread_title[0]) {
1907                               /*
1908                                * Args set via writable user thread mmap or
1909                                * sysctl().
1910                                *
1911                                * We must calculate the string length manually
1912                                * because the user data can change at any time.
1913                                */
1914                               size_t n;
1915                               char *base;
1916 
1917                               base = lp->lwp_lpmap->thread_title;
1918                               for (n = 0; n < LPMAP_MAXTHREADTITLE - 1; ++n) {
1919                                         if (base[n] == 0)
1920                                                   break;
1921                               }
1922                               error = SYSCTL_OUT(req, base, n);
1923                               if (error == 0)
1924                                         error = SYSCTL_OUT(req, "", 1);
1925                     } else if (p->p_upmap != NULL && p->p_upmap->proc_title[0]) {
1926                               /*
1927                                * Args set via writable user process mmap or
1928                                * sysctl().
1929                                *
1930                                * We must calculate the string length manually
1931                                * because the user data can change at any time.
1932                                */
1933                               size_t n;
1934                               char *base;
1935 
1936                               base = p->p_upmap->proc_title;
1937                               for (n = 0; n < UPMAP_MAXPROCTITLE - 1; ++n) {
1938                                         if (base[n] == 0)
1939                                                   break;
1940                               }
1941                               error = SYSCTL_OUT(req, base, n);
1942                               if (error == 0)
1943                                         error = SYSCTL_OUT(req, "", 1);
1944                     } else if ((pa = p->p_args) != NULL) {
1945                               /*
1946                                * Default/original arguments.
1947                                */
1948                               refcount_acquire(&pa->ar_ref);
1949                               error = SYSCTL_OUT(req, pa->ar_args, pa->ar_length);
1950                               if (refcount_release(&pa->ar_ref))
1951                                         kfree(pa, M_PARGS);
1952                     }
1953           }
1954           if (req->newptr == NULL)
1955                     goto done;
1956 
1957           if (req->newlen + sizeof(struct pargs) > ps_arg_cache_limit) {
1958                     goto done;
1959           }
1960 
1961           /*
1962            * Get the new process or thread title from userland
1963            */
1964           pa = kmalloc(sizeof(struct pargs) + req->newlen,
1965                          M_PARGS, M_WAITOK);
1966           refcount_init(&pa->ar_ref, 1);
1967           pa->ar_length = req->newlen;
1968           error = SYSCTL_IN(req, pa->ar_args, req->newlen);
1969           if (error) {
1970                     kfree(pa, M_PARGS);
1971                     goto done;
1972           }
1973 
1974           if (lp) {
1975                     /*
1976                      * Update thread title
1977                      */
1978                     if (lp->lwp_lpmap == NULL)
1979                               lwp_usermap(lp, -1);
1980                     if (lp->lwp_lpmap) {
1981                               n = req->newlen;
1982                               if (n >= sizeof(lp->lwp_lpmap->thread_title))
1983                                         n = sizeof(lp->lwp_lpmap->thread_title) - 1;
1984                               lp->lwp_lpmap->thread_title[n] = 0;
1985                               bcopy(pa->ar_args, lp->lwp_lpmap->thread_title, n);
1986                     }
1987           } else {
1988                     /*
1989                      * Update process title
1990                      */
1991                     if (p->p_upmap == NULL)
1992                               proc_usermap(p, -1);
1993                     if (p->p_upmap) {
1994                               n = req->newlen;
1995                               if (n >= sizeof(lp->lwp_lpmap->thread_title))
1996                                         n = sizeof(lp->lwp_lpmap->thread_title) - 1;
1997                               p->p_upmap->proc_title[n] = 0;
1998                               bcopy(pa->ar_args, p->p_upmap->proc_title, n);
1999                     }
2000 
2001 #if 0
2002                     /*
2003                      * XXX delete this code, keep original args intact for
2004                      * the setproctitle("") case.
2005                      * Scrap p->p_args, p->p_upmap->proc_title[] overrides it.
2006                      */
2007                     opa = p->p_args;
2008                     p->p_args = NULL;
2009                     if (opa) {
2010                               KKASSERT(opa->ar_ref > 0);
2011                               if (refcount_release(&opa->ar_ref)) {
2012                                         kfree(opa, M_PARGS);
2013                                         /* opa = NULL; */
2014                               }
2015                     }
2016 #endif
2017           }
2018           kfree(pa, M_PARGS);
2019 
2020 done:
2021           if (lp) {
2022                     lwkt_reltoken(&lp->lwp_token);
2023                     LWPRELE(lp);
2024           }
2025           if (p) {
2026                     lwkt_reltoken(&p->p_token);
2027                     PRELE(p);
2028           }
2029           return (error);
2030 }
2031 
2032 static int
sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)2033 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS)
2034 {
2035           int *name = (int*) arg1;
2036           u_int namelen = arg2;
2037           struct proc *p;
2038           int error = 0;
2039           char *fullpath, *freepath;
2040           struct ucred *cr1 = curproc->p_ucred;
2041 
2042           if (namelen != 1)
2043                     return (EINVAL);
2044 
2045           p = pfind((pid_t)name[0]);
2046           if (p == NULL)
2047                     goto done;
2048           lwkt_gettoken_shared(&p->p_token);
2049 
2050           /*
2051            * If we are not allowed to see other args, we certainly shouldn't
2052            * get the cwd either. Also check the usual trespassing.
2053            */
2054           if ((!ps_argsopen) && p_trespass(cr1, p->p_ucred))
2055                     goto done;
2056 
2057           if (req->oldptr && p->p_fd != NULL && p->p_fd->fd_ncdir.ncp) {
2058                     struct nchandle nch;
2059 
2060                     cache_copy(&p->p_fd->fd_ncdir, &nch);
2061                     error = cache_fullpath(p, &nch, NULL,
2062                                                &fullpath, &freepath, 0);
2063                     cache_drop(&nch);
2064                     if (error)
2065                               goto done;
2066                     error = SYSCTL_OUT(req, fullpath, strlen(fullpath) + 1);
2067                     kfree(freepath, M_TEMP);
2068           }
2069 
2070 done:
2071           if (p) {
2072                     lwkt_reltoken(&p->p_token);
2073                     PRELE(p);
2074           }
2075           return (error);
2076 }
2077 
2078 /*
2079  * This sysctl allows a process to retrieve the path of the executable for
2080  * itself or another process.
2081  */
2082 static int
sysctl_kern_proc_pathname(SYSCTL_HANDLER_ARGS)2083 sysctl_kern_proc_pathname(SYSCTL_HANDLER_ARGS)
2084 {
2085           pid_t *pidp = (pid_t *)arg1;
2086           unsigned int arglen = arg2;
2087           struct proc *p;
2088           char *retbuf, *freebuf;
2089           int error = 0;
2090           struct nchandle nch;
2091 
2092           if (arglen != 1)
2093                     return (EINVAL);
2094           if (*pidp == -1) {  /* -1 means this process */
2095                     p = curproc;
2096           } else {
2097                     p = pfind(*pidp);
2098                     if (p == NULL)
2099                               return (ESRCH);
2100           }
2101           lwkt_gettoken_shared(&p->p_token);      /* deal with exit race */
2102           if (p->p_textnch.ncp) {
2103                     cache_copy(&p->p_textnch, &nch);
2104                     error = cache_fullpath(p, &nch, NULL, &retbuf, &freebuf, 0);
2105                     cache_drop(&nch);
2106           } else {
2107                     error = EINVAL;
2108           }
2109           lwkt_reltoken(&p->p_token);
2110           if (error)
2111                     goto done;
2112           error = SYSCTL_OUT(req, retbuf, strlen(retbuf) + 1);
2113           kfree(freebuf, M_TEMP);
2114 done:
2115           if (*pidp != -1)
2116                     PRELE(p);
2117 
2118           return (error);
2119 }
2120 
2121 static int
sysctl_kern_proc_sigtramp(SYSCTL_HANDLER_ARGS)2122 sysctl_kern_proc_sigtramp(SYSCTL_HANDLER_ARGS)
2123 {
2124         /*int *name = (int *)arg1;*/
2125         u_int namelen = arg2;
2126         struct kinfo_sigtramp kst;
2127         const struct sysentvec *sv;
2128         int error;
2129 
2130         if (namelen > 1)
2131                 return (EINVAL);
2132         /* ignore pid if passed in (freebsd compatibility) */
2133 
2134         sv = curproc->p_sysent;
2135         bzero(&kst, sizeof(kst));
2136         if (sv->sv_szsigcode) {
2137                     intptr_t sigbase;
2138 
2139                     sigbase = trunc_page64((intptr_t)PS_STRINGS -
2140                                                *sv->sv_szsigcode);
2141                     sigbase -= SZSIGCODE_EXTRA_BYTES;
2142 
2143                 kst.ksigtramp_start = (void *)sigbase;
2144                 kst.ksigtramp_end = (void *)(sigbase + *sv->sv_szsigcode);
2145         }
2146         error = SYSCTL_OUT(req, &kst, sizeof(kst));
2147 
2148         return (error);
2149 }
2150 
2151 SYSCTL_NODE(_kern, KERN_PROC, proc, CTLFLAG_RD,  0, "Process table");
2152 
2153 #define SYSCTL_KERN_PROC_ALLFLAGS(which, affix)                                 \
2154     SYSCTL_NODE(_kern_proc,                                                     \
2155           which,                                                                          \
2156           affix,                                                                          \
2157           CTLFLAG_RD | CTLFLAG_NOLOCK,                                          \
2158           sysctl_kern_proc, "Process Table");                                   \
2159     SYSCTL_NODE(_kern_proc,                                                     \
2160           (which | KERN_PROC_FLAG_LWP),                                         \
2161           affix ## _lwp,                                                                  \
2162           CTLFLAG_RD | CTLFLAG_NOLOCK,                                          \
2163           sysctl_kern_proc, "Process Table");                                   \
2164     SYSCTL_NODE(_kern_proc,                                                     \
2165           (which | KERN_PROC_FLAG_LWKT),                                                  \
2166           affix ## _lwkt,                                                                 \
2167           CTLFLAG_RD | CTLFLAG_NOLOCK,                                          \
2168           sysctl_kern_proc, "Process Table");                                   \
2169     SYSCTL_NODE(_kern_proc,                                                     \
2170           (which | KERN_PROC_FLAG_LWP | KERN_PROC_FLAG_LWKT),                   \
2171           affix ## _lwp_lwkt,                                                   \
2172           CTLFLAG_RD | CTLFLAG_NOLOCK,                                          \
2173           sysctl_kern_proc, "Process Table")
2174 
2175 SYSCTL_PROC(_kern_proc,
2176           KERN_PROC_ALL,
2177           all,
2178           CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_NOLOCK,
2179           0, 0, sysctl_kern_proc, "S,proc", "Return entire process table");
2180 
2181 SYSCTL_NODE(_kern_proc,
2182           (KERN_PROC_ALL | KERN_PROC_FLAG_LWP),
2183           all_lwp,
2184           CTLFLAG_RD | CTLFLAG_NOLOCK,
2185           sysctl_kern_proc, "Process table");
2186 
2187 SYSCTL_NODE(_kern_proc,
2188           (KERN_PROC_ALL | KERN_PROC_FLAG_LWKT),
2189           all_lwkt,
2190           CTLFLAG_RD | CTLFLAG_NOLOCK,
2191           sysctl_kern_proc, "Process table");
2192 
2193 SYSCTL_NODE(_kern_proc,
2194           (KERN_PROC_ALL | KERN_PROC_FLAG_LWP | KERN_PROC_FLAG_LWKT),
2195           all_lwp_lwkt,
2196           CTLFLAG_RD | CTLFLAG_NOLOCK,
2197           sysctl_kern_proc, "Process table");
2198 
2199 SYSCTL_KERN_PROC_ALLFLAGS(KERN_PROC_PGRP, pgrp);
2200 SYSCTL_KERN_PROC_ALLFLAGS(KERN_PROC_TTY, tty);
2201 SYSCTL_KERN_PROC_ALLFLAGS(KERN_PROC_UID, uid);
2202 SYSCTL_KERN_PROC_ALLFLAGS(KERN_PROC_RUID, ruid);
2203 SYSCTL_KERN_PROC_ALLFLAGS(KERN_PROC_PID, pid);
2204 
2205 
2206 SYSCTL_NODE(_kern_proc, KERN_PROC_ARGS, args,
2207           CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK,
2208           sysctl_kern_proc_args, "Process argument list");
2209 
2210 SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd,
2211           CTLFLAG_RD | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK,
2212           sysctl_kern_proc_cwd, "Process argument list");
2213 
2214 static SYSCTL_NODE(_kern_proc, KERN_PROC_PATHNAME, pathname,
2215           CTLFLAG_RD | CTLFLAG_NOLOCK,
2216           sysctl_kern_proc_pathname, "Process executable path");
2217 
2218 SYSCTL_PROC(_kern_proc, KERN_PROC_SIGTRAMP, sigtramp,
2219           CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_NOLOCK,
2220         0, 0, sysctl_kern_proc_sigtramp, "S,sigtramp",
2221         "Return sigtramp address range");
2222