2 * Copyright (c) 2009 Robert N. M. Watson
5 * This software was developed at the University of Cambridge Computer
6 * Laboratory with support from a grant from Google, Inc.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR 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 AUTHOR 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
31 * FreeBSD process descriptor facility.
33 * Some processes are represented by a file descriptor, which will be used in
34 * preference to signaling and pids for the purposes of process management,
35 * and is, in effect, a form of capability. When a process descriptor is
36 * used with a process, it ceases to be visible to certain traditional UNIX
37 * process facilities, such as waitpid(2).
41 * - At most one process descriptor will exist for any process, although
42 * references to that descriptor may be held from many processes (or even
43 * be in flight between processes over a local domain socket).
44 * - Last close on the process descriptor will terminate the process using
45 * SIGKILL and reparent it to init so that there's a process to reap it
46 * when it's done exiting.
47 * - If the process exits before the descriptor is closed, it will not
48 * generate SIGCHLD on termination, or be picked up by waitpid().
49 * - The pdkill(2) system call may be used to deliver a signal to the process
50 * using its process descriptor.
51 * - The pdwait4(2) system call may be used to block (or not) on a process
52 * descriptor to collect termination information.
56 * - How to handle ptrace(2)?
57 * - Will we want to add a pidtoprocdesc(2) system call to allow process
58 * descriptors to be created for processes without pdfork(2)?
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
64 #include <sys/param.h>
65 #include <sys/capsicum.h>
66 #include <sys/fcntl.h>
68 #include <sys/filedesc.h>
69 #include <sys/kernel.h>
71 #include <sys/mutex.h>
74 #include <sys/procdesc.h>
75 #include <sys/resourcevar.h>
77 #include <sys/sysproto.h>
78 #include <sys/sysctl.h>
79 #include <sys/systm.h>
80 #include <sys/ucred.h>
82 #include <security/audit/audit.h>
86 FEATURE(process_descriptors, "Process Descriptors");
88 static uma_zone_t procdesc_zone;
90 static fo_poll_t procdesc_poll;
91 static fo_kqfilter_t procdesc_kqfilter;
92 static fo_stat_t procdesc_stat;
93 static fo_close_t procdesc_close;
95 static struct fileops procdesc_ops = {
96 .fo_read = invfo_rdwr,
97 .fo_write = invfo_rdwr,
98 .fo_truncate = invfo_truncate,
99 .fo_ioctl = invfo_ioctl,
100 .fo_poll = procdesc_poll,
101 .fo_kqfilter = procdesc_kqfilter,
102 .fo_stat = procdesc_stat,
103 .fo_close = procdesc_close,
104 .fo_chmod = invfo_chmod,
105 .fo_chown = invfo_chown,
106 .fo_sendfile = invfo_sendfile,
107 .fo_flags = DFLAG_PASSABLE,
111 * Initialize with VFS so that process descriptors are available along with
112 * other file descriptor types. As long as it runs before init(8) starts,
113 * there shouldn't be a problem.
116 procdesc_init(void *dummy __unused)
119 procdesc_zone = uma_zcreate("procdesc", sizeof(struct procdesc),
120 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
121 if (procdesc_zone == NULL)
122 panic("procdesc_init: procdesc_zone not initialized");
124 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL);
127 * Return a locked process given a process descriptor, or ESRCH if it has
131 procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp,
138 error = fget(td, fd, rightsp, &fp);
141 if (fp->f_type != DTYPE_PROCDESC) {
146 sx_slock(&proctree_lock);
147 if (pd->pd_proc != NULL) {
152 sx_sunlock(&proctree_lock);
159 * Function to be used by procstat(1) sysctls when returning procdesc
163 procdesc_pid(struct file *fp_procdesc)
167 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC,
168 ("procdesc_pid: !procdesc"));
170 pd = fp_procdesc->f_data;
175 * Retrieve the PID associated with a process descriptor.
178 kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp)
183 error = fget(td, fd, rightsp, &fp);
186 if (fp->f_type != DTYPE_PROCDESC) {
190 *pidp = procdesc_pid(fp);
197 * System call to return the pid of a process given its process descriptor.
200 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
206 AUDIT_ARG_FD(uap->fd);
207 error = kern_pdgetpid(td, uap->fd,
208 cap_rights_init(&rights, CAP_PDGETPID), &pid);
210 error = copyout(&pid, uap->pidp, sizeof(pid));
215 * When a new process is forked by pdfork(), a file descriptor is allocated
216 * by the fork code first, then the process is forked, and then we get a
217 * chance to set up the process descriptor. Failure is not permitted at this
218 * point, so procdesc_new() must succeed.
221 procdesc_new(struct proc *p, int flags)
225 pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO);
227 pd->pd_pid = p->p_pid;
230 if (flags & PD_DAEMON)
231 pd->pd_flags |= PDF_DAEMON;
232 PROCDESC_LOCK_INIT(pd);
233 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock);
236 * Process descriptors start out with two references: one from their
237 * struct file, and the other from their struct proc.
239 refcount_init(&pd->pd_refcount, 2);
243 * Initialize a file with a process descriptor.
246 procdesc_finit(struct procdesc *pdp, struct file *fp)
249 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
253 procdesc_free(struct procdesc *pd)
257 * When the last reference is released, we assert that the descriptor
258 * has been closed, but not that the process has exited, as we will
259 * detach the descriptor before the process dies if the descript is
260 * closed, as we can't wait synchronously.
262 if (refcount_release(&pd->pd_refcount)) {
263 KASSERT(pd->pd_proc == NULL,
264 ("procdesc_free: pd_proc != NULL"));
265 KASSERT((pd->pd_flags & PDF_CLOSED),
266 ("procdesc_free: !PDF_CLOSED"));
268 knlist_destroy(&pd->pd_selinfo.si_note);
269 PROCDESC_LOCK_DESTROY(pd);
270 uma_zfree(procdesc_zone, pd);
275 * procdesc_exit() - notify a process descriptor that its process is exiting.
276 * We use the proctree_lock to ensure that process exit either happens
277 * strictly before or strictly after a concurrent call to procdesc_close().
280 procdesc_exit(struct proc *p)
284 sx_assert(&proctree_lock, SA_XLOCKED);
285 PROC_LOCK_ASSERT(p, MA_OWNED);
286 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL"));
291 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc,
292 ("procdesc_exit: closed && parent not init"));
294 pd->pd_flags |= PDF_EXITED;
295 pd->pd_xstat = p->p_xstat;
298 * If the process descriptor has been closed, then we have nothing
299 * to do; return 1 so that init will get SIGCHLD and do the reaping.
300 * Clean up the procdesc now rather than letting it happen during
303 if (pd->pd_flags & PDF_CLOSED) {
306 p->p_procdesc = NULL;
310 if (pd->pd_flags & PDF_SELECTED) {
311 pd->pd_flags &= ~PDF_SELECTED;
312 selwakeup(&pd->pd_selinfo);
314 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT);
320 * When a process descriptor is reaped, perhaps as a result of close() or
321 * pdwait4(), release the process's reference on the process descriptor.
324 procdesc_reap(struct proc *p)
328 sx_assert(&proctree_lock, SA_XLOCKED);
329 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
333 p->p_procdesc = NULL;
338 * procdesc_close() - last close on a process descriptor. If the process is
339 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let
340 * init(8) clean up the mess; if not, we have to clean up the zombie ourselves.
343 procdesc_close(struct file *fp, struct thread *td)
348 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc"));
351 fp->f_ops = &badfileops;
354 sx_xlock(&proctree_lock);
356 pd->pd_flags |= PDF_CLOSED;
361 * This is the case where process' exit status was already
362 * collected and procdesc_reap() was already called.
364 sx_xunlock(&proctree_lock);
367 if (p->p_state == PRS_ZOMBIE) {
369 * If the process is already dead and just awaiting
370 * reaping, do that now. This will release the
371 * process's reference to the process descriptor when it
372 * calls back into procdesc_reap().
375 proc_reap(curthread, p, NULL, 0);
378 * If the process is not yet dead, we need to kill it,
379 * but we can't wait around synchronously for it to go
380 * away, as that path leads to madness (and deadlocks).
381 * First, detach the process from its descriptor so that
382 * its exit status will be reported normally.
385 p->p_procdesc = NULL;
389 * Next, reparent it to init(8) so that there's someone
390 * to pick up the pieces; finally, terminate with
393 p->p_sigparent = SIGCHLD;
394 proc_reparent(p, initproc);
395 if ((pd->pd_flags & PDF_DAEMON) == 0)
396 kern_psignal(p, SIGKILL);
398 sx_xunlock(&proctree_lock);
403 * Release the file descriptor's reference on the process descriptor.
410 procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
419 if (pd->pd_flags & PDF_EXITED)
422 selrecord(td, &pd->pd_selinfo);
423 pd->pd_flags |= PDF_SELECTED;
430 procdesc_kqops_detach(struct knote *kn)
434 pd = kn->kn_fp->f_data;
435 knlist_remove(&pd->pd_selinfo.si_note, kn, 0);
439 procdesc_kqops_event(struct knote *kn, long hint)
444 pd = kn->kn_fp->f_data;
447 * Initial test after registration. Generate a NOTE_EXIT in
448 * case the process already terminated before registration.
450 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0;
452 /* Mask off extra data. */
453 event = (u_int)hint & NOTE_PCTRLMASK;
456 /* If the user is interested in this event, record it. */
457 if (kn->kn_sfflags & event)
458 kn->kn_fflags |= event;
460 /* Process is gone, so flag the event as finished. */
461 if (event == NOTE_EXIT) {
462 kn->kn_flags |= EV_EOF | EV_ONESHOT;
463 if (kn->kn_fflags & NOTE_EXIT)
464 kn->kn_data = pd->pd_xstat;
465 if (kn->kn_fflags == 0)
466 kn->kn_flags |= EV_DROP;
470 return (kn->kn_fflags != 0);
473 static struct filterops procdesc_kqops = {
475 .f_detach = procdesc_kqops_detach,
476 .f_event = procdesc_kqops_event,
480 procdesc_kqfilter(struct file *fp, struct knote *kn)
485 switch (kn->kn_filter) {
486 case EVFILT_PROCDESC:
487 kn->kn_fop = &procdesc_kqops;
488 kn->kn_flags |= EV_CLEAR;
489 knlist_add(&pd->pd_selinfo.si_note, kn, 0);
497 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
501 struct timeval pstart;
504 * XXXRW: Perhaps we should cache some more information from the
505 * process so that we can return it reliably here even after it has
506 * died. For example, caching its credential data.
508 bzero(sb, sizeof(*sb));
510 sx_slock(&proctree_lock);
511 if (pd->pd_proc != NULL) {
512 PROC_LOCK(pd->pd_proc);
514 /* Set birth and [acm] times to process start time. */
515 pstart = pd->pd_proc->p_stats->p_start;
516 timevaladd(&pstart, &boottime);
517 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim);
518 sb->st_atim = sb->st_birthtim;
519 sb->st_ctim = sb->st_birthtim;
520 sb->st_mtim = sb->st_birthtim;
521 if (pd->pd_proc->p_state != PRS_ZOMBIE)
522 sb->st_mode = S_IFREG | S_IRWXU;
524 sb->st_mode = S_IFREG;
525 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid;
526 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid;
527 PROC_UNLOCK(pd->pd_proc);
529 sb->st_mode = S_IFREG;
530 sx_sunlock(&proctree_lock);