1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2009, 2016 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * This software was developed at the University of Cambridge Computer
8 * Laboratory with support from a grant from Google, Inc.
9 *
10 * Portions of this software were developed by BAE Systems, the University of
11 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
12 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
13 * Computing (TC) research program.
14 *
15 * Redistribution and use in source and binary forms, with or without
16 * modification, are permitted provided that the following conditions
17 * are met:
18 * 1. Redistributions of source code must retain the above copyright
19 * notice, this list of conditions and the following disclaimer.
20 * 2. Redistributions in binary form must reproduce the above copyright
21 * notice, this list of conditions and the following disclaimer in the
22 * documentation and/or other materials provided with the distribution.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36
37 /*-
38 * FreeBSD process descriptor facility.
39 *
40 * Some processes are represented by a file descriptor, which will be used in
41 * preference to signaling and pids for the purposes of process management,
42 * and is, in effect, a form of capability. When a process descriptor is
43 * used with a process, it ceases to be visible to certain traditional UNIX
44 * process facilities, such as waitpid(2).
45 *
46 * Some semantics:
47 *
48 * - At most one process descriptor will exist for any process, although
49 * references to that descriptor may be held from many processes (or even
50 * be in flight between processes over a local domain socket).
51 * - Last close on the process descriptor will terminate the process using
52 * SIGKILL and reparent it to init so that there's a process to reap it
53 * when it's done exiting.
54 * - If the process exits before the descriptor is closed, it will not
55 * generate SIGCHLD on termination, or be picked up by waitpid().
56 * - The pdkill(2) system call may be used to deliver a signal to the process
57 * using its process descriptor.
58 *
59 * Open questions:
60 *
61 * - Will we want to add a pidtoprocdesc(2) system call to allow process
62 * descriptors to be created for processes without pdfork(2)?
63 */
64
65 #include <sys/param.h>
66 #include <sys/capsicum.h>
67 #include <sys/fcntl.h>
68 #include <sys/file.h>
69 #include <sys/filedesc.h>
70 #include <sys/kernel.h>
71 #include <sys/lock.h>
72 #include <sys/mutex.h>
73 #include <sys/poll.h>
74 #include <sys/proc.h>
75 #include <sys/procdesc.h>
76 #include <sys/resourcevar.h>
77 #include <sys/stat.h>
78 #include <sys/syscallsubr.h>
79 #include <sys/sysproto.h>
80 #include <sys/sysctl.h>
81 #include <sys/systm.h>
82 #include <sys/ucred.h>
83 #include <sys/user.h>
84
85 #include <security/audit/audit.h>
86
87 #include <vm/uma.h>
88
89 FEATURE(process_descriptors, "Process Descriptors");
90
91 MALLOC_DEFINE(M_PROCDESC, "procdesc", "process descriptors");
92
93 static fo_poll_t procdesc_poll;
94 static fo_kqfilter_t procdesc_kqfilter;
95 static fo_stat_t procdesc_stat;
96 static fo_close_t procdesc_close;
97 static fo_fill_kinfo_t procdesc_fill_kinfo;
98 static fo_cmp_t procdesc_cmp;
99
100 static const struct fileops procdesc_ops = {
101 .fo_read = invfo_rdwr,
102 .fo_write = invfo_rdwr,
103 .fo_truncate = invfo_truncate,
104 .fo_ioctl = invfo_ioctl,
105 .fo_poll = procdesc_poll,
106 .fo_kqfilter = procdesc_kqfilter,
107 .fo_stat = procdesc_stat,
108 .fo_close = procdesc_close,
109 .fo_chmod = invfo_chmod,
110 .fo_chown = invfo_chown,
111 .fo_sendfile = invfo_sendfile,
112 .fo_fill_kinfo = procdesc_fill_kinfo,
113 .fo_cmp = procdesc_cmp,
114 .fo_flags = DFLAG_PASSABLE,
115 };
116
117 /*
118 * Return a locked process given a process descriptor, or ESRCH if it has
119 * died.
120 */
121 int
procdesc_find(struct thread * td,int fd,const cap_rights_t * rightsp,struct proc ** p)122 procdesc_find(struct thread *td, int fd, const cap_rights_t *rightsp,
123 struct proc **p)
124 {
125 struct procdesc *pd;
126 struct file *fp;
127 int error;
128
129 error = fget(td, fd, rightsp, &fp);
130 if (error)
131 return (error);
132 if (fp->f_type != DTYPE_PROCDESC) {
133 error = EINVAL;
134 goto out;
135 }
136 pd = fp->f_data;
137 sx_slock(&proctree_lock);
138 if (pd->pd_proc != NULL) {
139 *p = pd->pd_proc;
140 PROC_LOCK(*p);
141 } else
142 error = ESRCH;
143 sx_sunlock(&proctree_lock);
144 out:
145 fdrop(fp, td);
146 return (error);
147 }
148
149 /*
150 * Function to be used by procstat(1) sysctls when returning procdesc
151 * information.
152 */
153 pid_t
procdesc_pid(struct file * fp_procdesc)154 procdesc_pid(struct file *fp_procdesc)
155 {
156 struct procdesc *pd;
157
158 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC,
159 ("procdesc_pid: !procdesc"));
160
161 pd = fp_procdesc->f_data;
162 return (pd->pd_pid);
163 }
164
165 /*
166 * Retrieve the PID associated with a process descriptor.
167 */
168 int
kern_pdgetpid(struct thread * td,int fd,const cap_rights_t * rightsp,pid_t * pidp)169 kern_pdgetpid(struct thread *td, int fd, const cap_rights_t *rightsp,
170 pid_t *pidp)
171 {
172 struct file *fp;
173 int error;
174
175 error = fget(td, fd, rightsp, &fp);
176 if (error)
177 return (error);
178 if (fp->f_type != DTYPE_PROCDESC) {
179 error = EBADF;
180 goto out;
181 }
182 *pidp = procdesc_pid(fp);
183 out:
184 fdrop(fp, td);
185 return (error);
186 }
187
188 /*
189 * System call to return the pid of a process given its process descriptor.
190 */
191 int
sys_pdgetpid(struct thread * td,struct pdgetpid_args * uap)192 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
193 {
194 pid_t pid;
195 int error;
196
197 AUDIT_ARG_FD(uap->fd);
198 error = kern_pdgetpid(td, uap->fd, &cap_pdgetpid_rights, &pid);
199 if (error == 0)
200 error = copyout(&pid, uap->pidp, sizeof(pid));
201 return (error);
202 }
203
204 /*
205 * When a new process is forked by pdfork(), a file descriptor is allocated
206 * by the fork code first, then the process is forked, and then we get a
207 * chance to set up the process descriptor. Failure is not permitted at this
208 * point, so procdesc_new() must succeed.
209 */
210 void
procdesc_new(struct proc * p,int flags)211 procdesc_new(struct proc *p, int flags)
212 {
213 struct procdesc *pd;
214
215 pd = malloc(sizeof(*pd), M_PROCDESC, M_WAITOK | M_ZERO);
216 pd->pd_proc = p;
217 pd->pd_pid = p->p_pid;
218 p->p_procdesc = pd;
219 pd->pd_flags = 0;
220 if (flags & PD_DAEMON)
221 pd->pd_flags |= PDF_DAEMON;
222 PROCDESC_LOCK_INIT(pd);
223 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock);
224
225 /*
226 * Process descriptors start out with two references: one from their
227 * struct file, and the other from their struct proc.
228 */
229 refcount_init(&pd->pd_refcount, 2);
230 }
231
232 /*
233 * Create a new process decriptor for the process that refers to it.
234 */
235 int
procdesc_falloc(struct thread * td,struct file ** resultfp,int * resultfd,int flags,struct filecaps * fcaps)236 procdesc_falloc(struct thread *td, struct file **resultfp, int *resultfd,
237 int flags, struct filecaps *fcaps)
238 {
239 int fflags;
240
241 fflags = 0;
242 if (flags & PD_CLOEXEC)
243 fflags = O_CLOEXEC;
244
245 return (falloc_caps(td, resultfp, resultfd, fflags, fcaps));
246 }
247
248 /*
249 * Initialize a file with a process descriptor.
250 */
251 void
procdesc_finit(struct procdesc * pdp,struct file * fp)252 procdesc_finit(struct procdesc *pdp, struct file *fp)
253 {
254
255 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
256 }
257
258 static void
procdesc_free(struct procdesc * pd)259 procdesc_free(struct procdesc *pd)
260 {
261
262 /*
263 * When the last reference is released, we assert that the descriptor
264 * has been closed, but not that the process has exited, as we will
265 * detach the descriptor before the process dies if the descript is
266 * closed, as we can't wait synchronously.
267 */
268 if (refcount_release(&pd->pd_refcount)) {
269 KASSERT(pd->pd_proc == NULL,
270 ("procdesc_free: pd_proc != NULL"));
271 KASSERT((pd->pd_flags & PDF_CLOSED),
272 ("procdesc_free: !PDF_CLOSED"));
273
274 if (pd->pd_pid != -1)
275 proc_id_clear(PROC_ID_PID, pd->pd_pid);
276
277 knlist_destroy(&pd->pd_selinfo.si_note);
278 PROCDESC_LOCK_DESTROY(pd);
279 free(pd, M_PROCDESC);
280 }
281 }
282
283 /*
284 * procdesc_exit() - notify a process descriptor that its process is exiting.
285 * We use the proctree_lock to ensure that process exit either happens
286 * strictly before or strictly after a concurrent call to procdesc_close().
287 */
288 int
procdesc_exit(struct proc * p)289 procdesc_exit(struct proc *p)
290 {
291 struct procdesc *pd;
292
293 sx_assert(&proctree_lock, SA_XLOCKED);
294 PROC_LOCK_ASSERT(p, MA_OWNED);
295 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL"));
296
297 pd = p->p_procdesc;
298
299 PROCDESC_LOCK(pd);
300 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == p->p_reaper,
301 ("procdesc_exit: closed && parent not reaper"));
302
303 pd->pd_flags |= PDF_EXITED;
304 pd->pd_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig);
305
306 /*
307 * If the process descriptor has been closed, then we have nothing
308 * to do; return 1 so that init will get SIGCHLD and do the reaping.
309 * Clean up the procdesc now rather than letting it happen during
310 * that reap.
311 */
312 if (pd->pd_flags & PDF_CLOSED) {
313 PROCDESC_UNLOCK(pd);
314 pd->pd_proc = NULL;
315 p->p_procdesc = NULL;
316 procdesc_free(pd);
317 return (1);
318 }
319 if (pd->pd_flags & PDF_SELECTED) {
320 pd->pd_flags &= ~PDF_SELECTED;
321 selwakeup(&pd->pd_selinfo);
322 }
323 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT);
324 PROCDESC_UNLOCK(pd);
325
326 /* Wakeup all waiters for this procdesc' process exit. */
327 wakeup(&p->p_procdesc);
328 return (0);
329 }
330
331 /*
332 * When a process descriptor is reaped, perhaps as a result of close(), release
333 * the process's reference on the process descriptor.
334 */
335 void
procdesc_reap(struct proc * p)336 procdesc_reap(struct proc *p)
337 {
338 struct procdesc *pd;
339
340 sx_assert(&proctree_lock, SA_XLOCKED);
341 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
342
343 pd = p->p_procdesc;
344 pd->pd_proc = NULL;
345 p->p_procdesc = NULL;
346 procdesc_free(pd);
347 }
348
349 /*
350 * procdesc_close() - last close on a process descriptor. If the process is
351 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let
352 * its reaper clean up the mess; if not, we have to clean up the zombie
353 * ourselves.
354 */
355 static int
procdesc_close(struct file * fp,struct thread * td)356 procdesc_close(struct file *fp, struct thread *td)
357 {
358 struct procdesc *pd;
359 struct proc *p;
360
361 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc"));
362
363 pd = fp->f_data;
364 fp->f_ops = &badfileops;
365 fp->f_data = NULL;
366
367 sx_xlock(&proctree_lock);
368 PROCDESC_LOCK(pd);
369 pd->pd_flags |= PDF_CLOSED;
370 PROCDESC_UNLOCK(pd);
371 p = pd->pd_proc;
372 if (p == NULL) {
373 /*
374 * This is the case where process' exit status was already
375 * collected and procdesc_reap() was already called.
376 */
377 sx_xunlock(&proctree_lock);
378 } else {
379 PROC_LOCK(p);
380 AUDIT_ARG_PROCESS(p);
381 if (p->p_state == PRS_ZOMBIE) {
382 /*
383 * If the process is already dead and just awaiting
384 * reaping, do that now. This will release the
385 * process's reference to the process descriptor when it
386 * calls back into procdesc_reap().
387 */
388 proc_reap(curthread, p, NULL, 0);
389 } else {
390 /*
391 * If the process is not yet dead, we need to kill it,
392 * but we can't wait around synchronously for it to go
393 * away, as that path leads to madness (and deadlocks).
394 * First, detach the process from its descriptor so that
395 * its exit status will be reported normally.
396 */
397 pd->pd_proc = NULL;
398 p->p_procdesc = NULL;
399 pd->pd_pid = -1;
400 procdesc_free(pd);
401
402 /*
403 * Next, reparent it to its reaper (usually init(8)) so
404 * that there's someone to pick up the pieces; finally,
405 * terminate with prejudice.
406 */
407 p->p_sigparent = SIGCHLD;
408 if ((p->p_flag & P_TRACED) == 0) {
409 proc_reparent(p, p->p_reaper, true);
410 } else {
411 proc_clear_orphan(p);
412 p->p_oppid = p->p_reaper->p_pid;
413 proc_add_orphan(p, p->p_reaper);
414 }
415 if ((pd->pd_flags & PDF_DAEMON) == 0)
416 kern_psignal(p, SIGKILL);
417 PROC_UNLOCK(p);
418 sx_xunlock(&proctree_lock);
419 }
420 }
421
422 /*
423 * Release the file descriptor's reference on the process descriptor.
424 */
425 procdesc_free(pd);
426 return (0);
427 }
428
429 static int
procdesc_poll(struct file * fp,int events,struct ucred * active_cred,struct thread * td)430 procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
431 struct thread *td)
432 {
433 struct procdesc *pd;
434 int revents;
435
436 revents = 0;
437 pd = fp->f_data;
438 PROCDESC_LOCK(pd);
439 if (pd->pd_flags & PDF_EXITED)
440 revents |= POLLHUP;
441 if (revents == 0) {
442 selrecord(td, &pd->pd_selinfo);
443 pd->pd_flags |= PDF_SELECTED;
444 }
445 PROCDESC_UNLOCK(pd);
446 return (revents);
447 }
448
449 static void
procdesc_kqops_detach(struct knote * kn)450 procdesc_kqops_detach(struct knote *kn)
451 {
452 struct procdesc *pd;
453
454 pd = kn->kn_fp->f_data;
455 knlist_remove(&pd->pd_selinfo.si_note, kn, 0);
456 }
457
458 static int
procdesc_kqops_event(struct knote * kn,long hint)459 procdesc_kqops_event(struct knote *kn, long hint)
460 {
461 struct procdesc *pd;
462 u_int event;
463
464 pd = kn->kn_fp->f_data;
465 if (hint == 0) {
466 /*
467 * Initial test after registration. Generate a NOTE_EXIT in
468 * case the process already terminated before registration.
469 */
470 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0;
471 } else {
472 /* Mask off extra data. */
473 event = (u_int)hint & NOTE_PCTRLMASK;
474 }
475
476 /* If the user is interested in this event, record it. */
477 if (kn->kn_sfflags & event)
478 kn->kn_fflags |= event;
479
480 /* Process is gone, so flag the event as finished. */
481 if (event == NOTE_EXIT) {
482 kn->kn_flags |= EV_EOF | EV_ONESHOT;
483 if (kn->kn_fflags & NOTE_EXIT)
484 kn->kn_data = pd->pd_xstat;
485 if (kn->kn_fflags == 0)
486 kn->kn_flags |= EV_DROP;
487 return (1);
488 }
489
490 return (kn->kn_fflags != 0);
491 }
492
493 static const struct filterops procdesc_kqops = {
494 .f_isfd = 1,
495 .f_detach = procdesc_kqops_detach,
496 .f_event = procdesc_kqops_event,
497 .f_copy = knote_triv_copy,
498 };
499
500 static int
procdesc_kqfilter(struct file * fp,struct knote * kn)501 procdesc_kqfilter(struct file *fp, struct knote *kn)
502 {
503 struct procdesc *pd;
504
505 pd = fp->f_data;
506 switch (kn->kn_filter) {
507 case EVFILT_PROCDESC:
508 kn->kn_fop = &procdesc_kqops;
509 kn->kn_flags |= EV_CLEAR;
510 knlist_add(&pd->pd_selinfo.si_note, kn, 0);
511 return (0);
512 default:
513 return (EINVAL);
514 }
515 }
516
517 static int
procdesc_stat(struct file * fp,struct stat * sb,struct ucred * active_cred)518 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
519 {
520 struct procdesc *pd;
521 struct timeval pstart, boottime;
522
523 /*
524 * XXXRW: Perhaps we should cache some more information from the
525 * process so that we can return it reliably here even after it has
526 * died. For example, caching its credential data.
527 */
528 bzero(sb, sizeof(*sb));
529 pd = fp->f_data;
530 sx_slock(&proctree_lock);
531 if (pd->pd_proc != NULL) {
532 PROC_LOCK(pd->pd_proc);
533 AUDIT_ARG_PROCESS(pd->pd_proc);
534
535 /* Set birth and [acm] times to process start time. */
536 pstart = pd->pd_proc->p_stats->p_start;
537 getboottime(&boottime);
538 timevaladd(&pstart, &boottime);
539 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim);
540 sb->st_atim = sb->st_birthtim;
541 sb->st_ctim = sb->st_birthtim;
542 sb->st_mtim = sb->st_birthtim;
543 if (pd->pd_proc->p_state != PRS_ZOMBIE)
544 sb->st_mode = S_IFREG | S_IRWXU;
545 else
546 sb->st_mode = S_IFREG;
547 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid;
548 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid;
549 PROC_UNLOCK(pd->pd_proc);
550 } else
551 sb->st_mode = S_IFREG;
552 sx_sunlock(&proctree_lock);
553 return (0);
554 }
555
556 static int
procdesc_fill_kinfo(struct file * fp,struct kinfo_file * kif,struct filedesc * fdp)557 procdesc_fill_kinfo(struct file *fp, struct kinfo_file *kif,
558 struct filedesc *fdp)
559 {
560 struct procdesc *pdp;
561
562 kif->kf_type = KF_TYPE_PROCDESC;
563 pdp = fp->f_data;
564 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;
565 return (0);
566 }
567
568 static int
procdesc_cmp(struct file * fp1,struct file * fp2,struct thread * td)569 procdesc_cmp(struct file *fp1, struct file *fp2, struct thread *td)
570 {
571 struct procdesc *pdp1, *pdp2;
572
573 if (fp2->f_type != DTYPE_PROCDESC)
574 return (3);
575 pdp1 = fp1->f_data;
576 pdp2 = fp2->f_data;
577 return (kcmp_cmp((uintptr_t)pdp1->pd_pid, (uintptr_t)pdp2->pd_pid));
578 }
579