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 * - How to handle ptrace(2)?
64 * - Will we want to add a pidtoprocdesc(2) system call to allow process
65 * descriptors to be created for processes without pdfork(2)?
68 #include <sys/cdefs.h>
69 __FBSDID("$FreeBSD$");
71 #include <sys/param.h>
72 #include <sys/capsicum.h>
73 #include <sys/fcntl.h>
75 #include <sys/filedesc.h>
76 #include <sys/kernel.h>
78 #include <sys/mutex.h>
81 #include <sys/procdesc.h>
82 #include <sys/resourcevar.h>
84 #include <sys/sysproto.h>
85 #include <sys/sysctl.h>
86 #include <sys/systm.h>
87 #include <sys/ucred.h>
90 #include <security/audit/audit.h>
94 FEATURE(process_descriptors, "Process Descriptors");
96 static uma_zone_t procdesc_zone;
98 static fo_poll_t procdesc_poll;
99 static fo_kqfilter_t procdesc_kqfilter;
100 static fo_stat_t procdesc_stat;
101 static fo_close_t procdesc_close;
102 static fo_fill_kinfo_t procdesc_fill_kinfo;
104 static struct fileops procdesc_ops = {
105 .fo_read = invfo_rdwr,
106 .fo_write = invfo_rdwr,
107 .fo_truncate = invfo_truncate,
108 .fo_ioctl = invfo_ioctl,
109 .fo_poll = procdesc_poll,
110 .fo_kqfilter = procdesc_kqfilter,
111 .fo_stat = procdesc_stat,
112 .fo_close = procdesc_close,
113 .fo_chmod = invfo_chmod,
114 .fo_chown = invfo_chown,
115 .fo_sendfile = invfo_sendfile,
116 .fo_fill_kinfo = procdesc_fill_kinfo,
117 .fo_flags = DFLAG_PASSABLE,
121 * Initialize with VFS so that process descriptors are available along with
122 * other file descriptor types. As long as it runs before init(8) starts,
123 * there shouldn't be a problem.
126 procdesc_init(void *dummy __unused)
129 procdesc_zone = uma_zcreate("procdesc", sizeof(struct procdesc),
130 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
131 if (procdesc_zone == NULL)
132 panic("procdesc_init: procdesc_zone not initialized");
134 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL);
137 * Return a locked process given a process descriptor, or ESRCH if it has
141 procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp,
148 error = fget(td, fd, rightsp, &fp);
151 if (fp->f_type != DTYPE_PROCDESC) {
156 sx_slock(&proctree_lock);
157 if (pd->pd_proc != NULL) {
162 sx_sunlock(&proctree_lock);
169 * Function to be used by procstat(1) sysctls when returning procdesc
173 procdesc_pid(struct file *fp_procdesc)
177 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC,
178 ("procdesc_pid: !procdesc"));
180 pd = fp_procdesc->f_data;
185 * Retrieve the PID associated with a process descriptor.
188 kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp)
193 error = fget(td, fd, rightsp, &fp);
196 if (fp->f_type != DTYPE_PROCDESC) {
200 *pidp = procdesc_pid(fp);
207 * System call to return the pid of a process given its process descriptor.
210 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
215 AUDIT_ARG_FD(uap->fd);
216 error = kern_pdgetpid(td, uap->fd, &cap_pdgetpid_rights, &pid);
218 error = copyout(&pid, uap->pidp, sizeof(pid));
223 * When a new process is forked by pdfork(), a file descriptor is allocated
224 * by the fork code first, then the process is forked, and then we get a
225 * chance to set up the process descriptor. Failure is not permitted at this
226 * point, so procdesc_new() must succeed.
229 procdesc_new(struct proc *p, int flags)
233 pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO);
235 pd->pd_pid = p->p_pid;
238 if (flags & PD_DAEMON)
239 pd->pd_flags |= PDF_DAEMON;
240 PROCDESC_LOCK_INIT(pd);
241 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock);
244 * Process descriptors start out with two references: one from their
245 * struct file, and the other from their struct proc.
247 refcount_init(&pd->pd_refcount, 2);
251 * Create a new process decriptor for the process that refers to it.
254 procdesc_falloc(struct thread *td, struct file **resultfp, int *resultfd,
255 int flags, struct filecaps *fcaps)
260 if (flags & PD_CLOEXEC)
263 return (falloc_caps(td, resultfp, resultfd, fflags, fcaps));
267 * Initialize a file with a process descriptor.
270 procdesc_finit(struct procdesc *pdp, struct file *fp)
273 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
277 procdesc_free(struct procdesc *pd)
281 * When the last reference is released, we assert that the descriptor
282 * has been closed, but not that the process has exited, as we will
283 * detach the descriptor before the process dies if the descript is
284 * closed, as we can't wait synchronously.
286 if (refcount_release(&pd->pd_refcount)) {
287 KASSERT(pd->pd_proc == NULL,
288 ("procdesc_free: pd_proc != NULL"));
289 KASSERT((pd->pd_flags & PDF_CLOSED),
290 ("procdesc_free: !PDF_CLOSED"));
292 knlist_destroy(&pd->pd_selinfo.si_note);
293 PROCDESC_LOCK_DESTROY(pd);
294 uma_zfree(procdesc_zone, pd);
299 * procdesc_exit() - notify a process descriptor that its process is exiting.
300 * We use the proctree_lock to ensure that process exit either happens
301 * strictly before or strictly after a concurrent call to procdesc_close().
304 procdesc_exit(struct proc *p)
308 sx_assert(&proctree_lock, SA_XLOCKED);
309 PROC_LOCK_ASSERT(p, MA_OWNED);
310 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL"));
315 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc,
316 ("procdesc_exit: closed && parent not init"));
318 pd->pd_flags |= PDF_EXITED;
319 pd->pd_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig);
322 * If the process descriptor has been closed, then we have nothing
323 * to do; return 1 so that init will get SIGCHLD and do the reaping.
324 * Clean up the procdesc now rather than letting it happen during
327 if (pd->pd_flags & PDF_CLOSED) {
330 p->p_procdesc = NULL;
334 if (pd->pd_flags & PDF_SELECTED) {
335 pd->pd_flags &= ~PDF_SELECTED;
336 selwakeup(&pd->pd_selinfo);
338 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT);
344 * When a process descriptor is reaped, perhaps as a result of close() or
345 * pdwait4(), release the process's reference on the process descriptor.
348 procdesc_reap(struct proc *p)
352 sx_assert(&proctree_lock, SA_XLOCKED);
353 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
357 p->p_procdesc = NULL;
362 * procdesc_close() - last close on a process descriptor. If the process is
363 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let
364 * init(8) clean up the mess; if not, we have to clean up the zombie ourselves.
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 init(8) so that there's someone
414 * to pick up the pieces; finally, terminate with
417 p->p_sigparent = SIGCHLD;
418 proc_reparent(p, initproc);
419 if ((pd->pd_flags & PDF_DAEMON) == 0)
420 kern_psignal(p, SIGKILL);
422 sx_xunlock(&proctree_lock);
427 * Release the file descriptor's reference on the process descriptor.
434 procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
443 if (pd->pd_flags & PDF_EXITED)
446 selrecord(td, &pd->pd_selinfo);
447 pd->pd_flags |= PDF_SELECTED;
454 procdesc_kqops_detach(struct knote *kn)
458 pd = kn->kn_fp->f_data;
459 knlist_remove(&pd->pd_selinfo.si_note, kn, 0);
463 procdesc_kqops_event(struct knote *kn, long hint)
468 pd = kn->kn_fp->f_data;
471 * Initial test after registration. Generate a NOTE_EXIT in
472 * case the process already terminated before registration.
474 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0;
476 /* Mask off extra data. */
477 event = (u_int)hint & NOTE_PCTRLMASK;
480 /* If the user is interested in this event, record it. */
481 if (kn->kn_sfflags & event)
482 kn->kn_fflags |= event;
484 /* Process is gone, so flag the event as finished. */
485 if (event == NOTE_EXIT) {
486 kn->kn_flags |= EV_EOF | EV_ONESHOT;
487 if (kn->kn_fflags & NOTE_EXIT)
488 kn->kn_data = pd->pd_xstat;
489 if (kn->kn_fflags == 0)
490 kn->kn_flags |= EV_DROP;
494 return (kn->kn_fflags != 0);
497 static struct filterops procdesc_kqops = {
499 .f_detach = procdesc_kqops_detach,
500 .f_event = procdesc_kqops_event,
504 procdesc_kqfilter(struct file *fp, struct knote *kn)
509 switch (kn->kn_filter) {
510 case EVFILT_PROCDESC:
511 kn->kn_fop = &procdesc_kqops;
512 kn->kn_flags |= EV_CLEAR;
513 knlist_add(&pd->pd_selinfo.si_note, kn, 0);
521 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
525 struct timeval pstart, boottime;
528 * XXXRW: Perhaps we should cache some more information from the
529 * process so that we can return it reliably here even after it has
530 * died. For example, caching its credential data.
532 bzero(sb, sizeof(*sb));
534 sx_slock(&proctree_lock);
535 if (pd->pd_proc != NULL) {
536 PROC_LOCK(pd->pd_proc);
537 AUDIT_ARG_PROCESS(pd->pd_proc);
539 /* Set birth and [acm] times to process start time. */
540 pstart = pd->pd_proc->p_stats->p_start;
541 getboottime(&boottime);
542 timevaladd(&pstart, &boottime);
543 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim);
544 sb->st_atim = sb->st_birthtim;
545 sb->st_ctim = sb->st_birthtim;
546 sb->st_mtim = sb->st_birthtim;
547 if (pd->pd_proc->p_state != PRS_ZOMBIE)
548 sb->st_mode = S_IFREG | S_IRWXU;
550 sb->st_mode = S_IFREG;
551 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid;
552 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid;
553 PROC_UNLOCK(pd->pd_proc);
555 sb->st_mode = S_IFREG;
556 sx_sunlock(&proctree_lock);
561 procdesc_fill_kinfo(struct file *fp, struct kinfo_file *kif,
562 struct filedesc *fdp)
564 struct procdesc *pdp;
566 kif->kf_type = KF_TYPE_PROCDESC;
568 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid;