2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2009, 2016 Robert N. M. Watson
7 * This software was developed at the University of Cambridge Computer
8 * Laboratory with support from a grant from Google, Inc.
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.
15 * Redistribution and use in source and binary forms, with or without
16 * modification, are permitted provided that the following conditions
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.
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
38 * FreeBSD process descriptor facility.
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).
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 * - The pdwait4(2) system call may be used to block (or not) on a process
59 * descriptor to collect termination information.
63 * - Will we want to add a pidtoprocdesc(2) system call to allow process
64 * descriptors to be created for processes without pdfork(2)?
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
70 #include <sys/param.h>
71 #include <sys/capsicum.h>
72 #include <sys/fcntl.h>
74 #include <sys/filedesc.h>
75 #include <sys/kernel.h>
77 #include <sys/mutex.h>
80 #include <sys/procdesc.h>
81 #include <sys/resourcevar.h>
83 #include <sys/sysproto.h>
84 #include <sys/sysctl.h>
85 #include <sys/systm.h>
86 #include <sys/ucred.h>
89 #include <security/audit/audit.h>
93 FEATURE(process_descriptors, "Process Descriptors");
95 static uma_zone_t procdesc_zone;
97 static fo_poll_t procdesc_poll;
98 static fo_kqfilter_t procdesc_kqfilter;
99 static fo_stat_t procdesc_stat;
100 static fo_close_t procdesc_close;
101 static fo_fill_kinfo_t procdesc_fill_kinfo;
103 static struct fileops procdesc_ops = {
104 .fo_read = invfo_rdwr,
105 .fo_write = invfo_rdwr,
106 .fo_truncate = invfo_truncate,
107 .fo_ioctl = invfo_ioctl,
108 .fo_poll = procdesc_poll,
109 .fo_kqfilter = procdesc_kqfilter,
110 .fo_stat = procdesc_stat,
111 .fo_close = procdesc_close,
112 .fo_chmod = invfo_chmod,
113 .fo_chown = invfo_chown,
114 .fo_sendfile = invfo_sendfile,
115 .fo_fill_kinfo = procdesc_fill_kinfo,
116 .fo_flags = DFLAG_PASSABLE,
120 * Initialize with VFS so that process descriptors are available along with
121 * other file descriptor types. As long as it runs before init(8) starts,
122 * there shouldn't be a problem.
125 procdesc_init(void *dummy __unused)
128 procdesc_zone = uma_zcreate("procdesc", sizeof(struct procdesc),
129 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
130 if (procdesc_zone == NULL)
131 panic("procdesc_init: procdesc_zone not initialized");
133 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL);
136 * Return a locked process given a process descriptor, or ESRCH if it has
140 procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp,
147 error = fget(td, fd, rightsp, &fp);
150 if (fp->f_type != DTYPE_PROCDESC) {
155 sx_slock(&proctree_lock);
156 if (pd->pd_proc != NULL) {
161 sx_sunlock(&proctree_lock);
168 * Function to be used by procstat(1) sysctls when returning procdesc
172 procdesc_pid(struct file *fp_procdesc)
176 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC,
177 ("procdesc_pid: !procdesc"));
179 pd = fp_procdesc->f_data;
184 * Retrieve the PID associated with a process descriptor.
187 kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp)
192 error = fget(td, fd, rightsp, &fp);
195 if (fp->f_type != DTYPE_PROCDESC) {
199 *pidp = procdesc_pid(fp);
206 * System call to return the pid of a process given its process descriptor.
209 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
214 AUDIT_ARG_FD(uap->fd);
215 error = kern_pdgetpid(td, uap->fd, &cap_pdgetpid_rights, &pid);
217 error = copyout(&pid, uap->pidp, sizeof(pid));
222 * When a new process is forked by pdfork(), a file descriptor is allocated
223 * by the fork code first, then the process is forked, and then we get a
224 * chance to set up the process descriptor. Failure is not permitted at this
225 * point, so procdesc_new() must succeed.
228 procdesc_new(struct proc *p, int flags)
232 pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO);
234 pd->pd_pid = p->p_pid;
237 if (flags & PD_DAEMON)
238 pd->pd_flags |= PDF_DAEMON;
239 PROCDESC_LOCK_INIT(pd);
240 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock);
243 * Process descriptors start out with two references: one from their
244 * struct file, and the other from their struct proc.
246 refcount_init(&pd->pd_refcount, 2);
250 * Create a new process decriptor for the process that refers to it.
253 procdesc_falloc(struct thread *td, struct file **resultfp, int *resultfd,
254 int flags, struct filecaps *fcaps)
259 if (flags & PD_CLOEXEC)
262 return (falloc_caps(td, resultfp, resultfd, fflags, fcaps));
266 * Initialize a file with a process descriptor.
269 procdesc_finit(struct procdesc *pdp, struct file *fp)
272 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
276 procdesc_free(struct procdesc *pd)
280 * When the last reference is released, we assert that the descriptor
281 * has been closed, but not that the process has exited, as we will
282 * detach the descriptor before the process dies if the descript is
283 * closed, as we can't wait synchronously.
285 if (refcount_release(&pd->pd_refcount)) {
286 KASSERT(pd->pd_proc == NULL,
287 ("procdesc_free: pd_proc != NULL"));
288 KASSERT((pd->pd_flags & PDF_CLOSED),
289 ("procdesc_free: !PDF_CLOSED"));
291 knlist_destroy(&pd->pd_selinfo.si_note);
292 PROCDESC_LOCK_DESTROY(pd);
293 uma_zfree(procdesc_zone, pd);
298 * procdesc_exit() - notify a process descriptor that its process is exiting.
299 * We use the proctree_lock to ensure that process exit either happens
300 * strictly before or strictly after a concurrent call to procdesc_close().
303 procdesc_exit(struct proc *p)
307 sx_assert(&proctree_lock, SA_XLOCKED);
308 PROC_LOCK_ASSERT(p, MA_OWNED);
309 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL"));
314 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == p->p_reaper,
315 ("procdesc_exit: closed && parent not reaper"));
317 pd->pd_flags |= PDF_EXITED;
318 pd->pd_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig);
321 * If the process descriptor has been closed, then we have nothing
322 * to do; return 1 so that init will get SIGCHLD and do the reaping.
323 * Clean up the procdesc now rather than letting it happen during
326 if (pd->pd_flags & PDF_CLOSED) {
329 p->p_procdesc = NULL;
333 if (pd->pd_flags & PDF_SELECTED) {
334 pd->pd_flags &= ~PDF_SELECTED;
335 selwakeup(&pd->pd_selinfo);
337 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT);
343 * When a process descriptor is reaped, perhaps as a result of close() or
344 * pdwait4(), release the process's reference on the process descriptor.
347 procdesc_reap(struct proc *p)
351 sx_assert(&proctree_lock, SA_XLOCKED);
352 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
356 p->p_procdesc = NULL;
361 * procdesc_close() - last close on a process descriptor. If the process is
362 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let
363 * its reaper clean up the mess; if not, we have to clean up the zombie
367 procdesc_close(struct file *fp, struct thread *td)
372 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc"));
375 fp->f_ops = &badfileops;
378 sx_xlock(&proctree_lock);
380 pd->pd_flags |= PDF_CLOSED;
385 * This is the case where process' exit status was already
386 * collected and procdesc_reap() was already called.
388 sx_xunlock(&proctree_lock);
391 AUDIT_ARG_PROCESS(p);
392 if (p->p_state == PRS_ZOMBIE) {
394 * If the process is already dead and just awaiting
395 * reaping, do that now. This will release the
396 * process's reference to the process descriptor when it
397 * calls back into procdesc_reap().
399 proc_reap(curthread, p, NULL, 0);
402 * If the process is not yet dead, we need to kill it,
403 * but we can't wait around synchronously for it to go
404 * away, as that path leads to madness (and deadlocks).
405 * First, detach the process from its descriptor so that
406 * its exit status will be reported normally.
409 p->p_procdesc = NULL;
413 * Next, reparent it to its reaper (usually init(8)) so
414 * that there's someone to pick up the pieces; finally,
415 * terminate with prejudice.
417 p->p_sigparent = SIGCHLD;
418 if ((p->p_flag & P_TRACED) == 0) {
419 proc_reparent(p, p->p_reaper, true);
421 proc_clear_orphan(p);
422 p->p_oppid = p->p_reaper->p_pid;
423 proc_add_orphan(p, p->p_reaper);
425 if ((pd->pd_flags & PDF_DAEMON) == 0)
426 kern_psignal(p, SIGKILL);
428 sx_xunlock(&proctree_lock);
433 * Release the file descriptor's reference on the process descriptor.
440 procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
449 if (pd->pd_flags & PDF_EXITED)
452 selrecord(td, &pd->pd_selinfo);
453 pd->pd_flags |= PDF_SELECTED;
460 procdesc_kqops_detach(struct knote *kn)
464 pd = kn->kn_fp->f_data;
465 knlist_remove(&pd->pd_selinfo.si_note, kn, 0);
469 procdesc_kqops_event(struct knote *kn, long hint)
474 pd = kn->kn_fp->f_data;
477 * Initial test after registration. Generate a NOTE_EXIT in
478 * case the process already terminated before registration.
480 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0;
482 /* Mask off extra data. */
483 event = (u_int)hint & NOTE_PCTRLMASK;
486 /* If the user is interested in this event, record it. */
487 if (kn->kn_sfflags & event)
488 kn->kn_fflags |= event;
490 /* Process is gone, so flag the event as finished. */
491 if (event == NOTE_EXIT) {
492 kn->kn_flags |= EV_EOF | EV_ONESHOT;
493 if (kn->kn_fflags & NOTE_EXIT)
494 kn->kn_data = pd->pd_xstat;
495 if (kn->kn_fflags == 0)
496 kn->kn_flags |= EV_DROP;
500 return (kn->kn_fflags != 0);
503 static struct filterops procdesc_kqops = {
505 .f_detach = procdesc_kqops_detach,
506 .f_event = procdesc_kqops_event,
510 procdesc_kqfilter(struct file *fp, struct knote *kn)
515 switch (kn->kn_filter) {
516 case EVFILT_PROCDESC:
517 kn->kn_fop = &procdesc_kqops;
518 kn->kn_flags |= EV_CLEAR;
519 knlist_add(&pd->pd_selinfo.si_note, kn, 0);
527 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
531 struct timeval pstart, boottime;
534 * XXXRW: Perhaps we should cache some more information from the
535 * process so that we can return it reliably here even after it has
536 * died. For example, caching its credential data.
538 bzero(sb, sizeof(*sb));
540 sx_slock(&proctree_lock);
541 if (pd->pd_proc != NULL) {
542 PROC_LOCK(pd->pd_proc);
543 AUDIT_ARG_PROCESS(pd->pd_proc);
545 /* Set birth and [acm] times to process start time. */
546 pstart = pd->pd_proc->p_stats->p_start;
547 getboottime(&boottime);
548 timevaladd(&pstart, &boottime);
549 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim);
550 sb->st_atim = sb->st_birthtim;
551 sb->st_ctim = sb->st_birthtim;
552 sb->st_mtim = sb->st_birthtim;
553 if (pd->pd_proc->p_state != PRS_ZOMBIE)
554 sb->st_mode = S_IFREG | S_IRWXU;
556 sb->st_mode = S_IFREG;
557 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid;
558 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid;
559 PROC_UNLOCK(pd->pd_proc);
561 sb->st_mode = S_IFREG;
562 sx_sunlock(&proctree_lock);
567 procdesc_fill_kinfo(struct file *fp, struct kinfo_file *kif,
568 struct filedesc *fdp)
570 struct procdesc *pdp;
572 kif->kf_type = KF_TYPE_PROCDESC;
574 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;