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 pfork(2)?
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
64 #include "opt_procdesc.h"
66 #include <sys/param.h>
67 #include <sys/capsicum.h>
68 #include <sys/fcntl.h>
70 #include <sys/filedesc.h>
71 #include <sys/kernel.h>
73 #include <sys/mutex.h>
76 #include <sys/procdesc.h>
77 #include <sys/resourcevar.h>
79 #include <sys/sysproto.h>
80 #include <sys/sysctl.h>
81 #include <sys/systm.h>
82 #include <sys/ucred.h>
84 #include <security/audit/audit.h>
90 FEATURE(process_descriptors, "Process Descriptors");
92 static uma_zone_t procdesc_zone;
94 static fo_rdwr_t procdesc_read;
95 static fo_rdwr_t procdesc_write;
96 static fo_truncate_t procdesc_truncate;
97 static fo_ioctl_t procdesc_ioctl;
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_chmod_t procdesc_chmod;
103 static fo_chown_t procdesc_chown;
105 static struct fileops procdesc_ops = {
106 .fo_read = procdesc_read,
107 .fo_write = procdesc_write,
108 .fo_truncate = procdesc_truncate,
109 .fo_ioctl = procdesc_ioctl,
110 .fo_poll = procdesc_poll,
111 .fo_kqfilter = procdesc_kqfilter,
112 .fo_stat = procdesc_stat,
113 .fo_close = procdesc_close,
114 .fo_chmod = procdesc_chmod,
115 .fo_chown = procdesc_chown,
116 .fo_sendfile = invfo_sendfile,
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)
216 AUDIT_ARG_FD(uap->fd);
217 error = kern_pdgetpid(td, uap->fd,
218 cap_rights_init(&rights, CAP_PDGETPID), &pid);
220 error = copyout(&pid, uap->pidp, sizeof(pid));
225 * When a new process is forked by pdfork(), a file descriptor is allocated
226 * by the fork code first, then the process is forked, and then we get a
227 * chance to set up the process descriptor. Failure is not permitted at this
228 * point, so procdesc_new() must succeed.
231 procdesc_new(struct proc *p, int flags)
235 pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO);
237 pd->pd_pid = p->p_pid;
240 if (flags & PD_DAEMON)
241 pd->pd_flags |= PDF_DAEMON;
242 PROCDESC_LOCK_INIT(pd);
245 * Process descriptors start out with two references: one from their
246 * struct file, and the other from their struct proc.
248 refcount_init(&pd->pd_refcount, 2);
252 * Initialize a file with a process descriptor.
255 procdesc_finit(struct procdesc *pdp, struct file *fp)
258 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
262 procdesc_free(struct procdesc *pd)
266 * When the last reference is released, we assert that the descriptor
267 * has been closed, but not that the process has exited, as we will
268 * detach the descriptor before the process dies if the descript is
269 * closed, as we can't wait synchronously.
271 if (refcount_release(&pd->pd_refcount)) {
272 KASSERT(pd->pd_proc == NULL,
273 ("procdesc_free: pd_proc != NULL"));
274 KASSERT((pd->pd_flags & PDF_CLOSED),
275 ("procdesc_free: !PDF_CLOSED"));
277 PROCDESC_LOCK_DESTROY(pd);
278 uma_zfree(procdesc_zone, pd);
283 * procdesc_exit() - notify a process descriptor that its process is exiting.
284 * We use the proctree_lock to ensure that process exit either happens
285 * strictly before or strictly after a concurrent call to procdesc_close().
288 procdesc_exit(struct proc *p)
292 sx_assert(&proctree_lock, SA_XLOCKED);
293 PROC_LOCK_ASSERT(p, MA_OWNED);
294 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL"));
299 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc,
300 ("procdesc_exit: closed && parent not init"));
302 pd->pd_flags |= PDF_EXITED;
305 * If the process descriptor has been closed, then we have nothing
306 * to do; return 1 so that init will get SIGCHLD and do the reaping.
307 * Clean up the procdesc now rather than letting it happen during
310 if (pd->pd_flags & PDF_CLOSED) {
313 p->p_procdesc = NULL;
317 if (pd->pd_flags & PDF_SELECTED) {
318 pd->pd_flags &= ~PDF_SELECTED;
319 selwakeup(&pd->pd_selinfo);
326 * When a process descriptor is reaped, perhaps as a result of close() or
327 * pdwait4(), release the process's reference on the process descriptor.
330 procdesc_reap(struct proc *p)
334 sx_assert(&proctree_lock, SA_XLOCKED);
335 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
339 p->p_procdesc = NULL;
344 * procdesc_close() - last close on a process descriptor. If the process is
345 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let
346 * init(8) clean up the mess; if not, we have to clean up the zombie ourselves.
349 procdesc_close(struct file *fp, struct thread *td)
354 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc"));
357 fp->f_ops = &badfileops;
360 sx_xlock(&proctree_lock);
362 pd->pd_flags |= PDF_CLOSED;
367 * This is the case where process' exit status was already
368 * collected and procdesc_reap() was already called.
370 sx_xunlock(&proctree_lock);
371 } else if (p->p_state == PRS_ZOMBIE) {
373 * If the process is already dead and just awaiting reaping,
374 * do that now. This will release the process's reference to
375 * the process descriptor when it calls back into
380 proc_reap(curthread, p, NULL, 0);
383 * If the process is not yet dead, we need to kill it, but we
384 * can't wait around synchronously for it to go away, as that
385 * path leads to madness (and deadlocks). First, detach the
386 * process from its descriptor so that its exit status will
387 * be reported normally.
391 p->p_procdesc = NULL;
395 * Next, reparent it to init(8) so that there's someone to
396 * pick up the pieces; finally, terminate with prejudice.
398 p->p_sigparent = SIGCHLD;
399 proc_reparent(p, initproc);
400 if ((pd->pd_flags & PDF_DAEMON) == 0)
401 kern_psignal(p, SIGKILL);
403 sx_xunlock(&proctree_lock);
407 * Release the file descriptor's reference on the process descriptor.
414 procdesc_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
415 int flags, struct thread *td)
422 procdesc_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
423 int flags, struct thread *td)
430 procdesc_truncate(struct file *fp, off_t length, struct ucred *active_cred,
438 procdesc_ioctl(struct file *fp, u_long com, void *data,
439 struct ucred *active_cred, struct thread *td)
446 procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
455 if (pd->pd_flags & PDF_EXITED)
458 selrecord(td, &pd->pd_selinfo);
459 pd->pd_flags |= PDF_SELECTED;
466 procdesc_kqfilter(struct file *fp, struct knote *kn)
473 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
477 struct timeval pstart;
480 * XXXRW: Perhaps we should cache some more information from the
481 * process so that we can return it reliably here even after it has
482 * died. For example, caching its credential data.
484 bzero(sb, sizeof(*sb));
486 sx_slock(&proctree_lock);
487 if (pd->pd_proc != NULL) {
488 PROC_LOCK(pd->pd_proc);
490 /* Set birth and [acm] times to process start time. */
491 pstart = pd->pd_proc->p_stats->p_start;
492 timevaladd(&pstart, &boottime);
493 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim);
494 sb->st_atim = sb->st_birthtim;
495 sb->st_ctim = sb->st_birthtim;
496 sb->st_mtim = sb->st_birthtim;
497 if (pd->pd_proc->p_state != PRS_ZOMBIE)
498 sb->st_mode = S_IFREG | S_IRWXU;
500 sb->st_mode = S_IFREG;
501 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid;
502 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid;
503 PROC_UNLOCK(pd->pd_proc);
505 sb->st_mode = S_IFREG;
506 sx_sunlock(&proctree_lock);
511 procdesc_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
519 procdesc_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
526 #else /* !PROCDESC */
529 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
535 #endif /* PROCDESC */