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/capability.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 rights,
148 error = fget(td, fd, rights, &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 rights, pid_t *pidp)
193 error = fget(td, fd, rights, &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, &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);
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 * Initialize a file with a process descriptor.
253 procdesc_finit(struct procdesc *pdp, struct file *fp)
256 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
260 procdesc_free(struct procdesc *pd)
264 * When the last reference is released, we assert that the descriptor
265 * has been closed, but not that the process has exited, as we will
266 * detach the descriptor before the process dies if the descript is
267 * closed, as we can't wait synchronously.
269 if (refcount_release(&pd->pd_refcount)) {
270 KASSERT(pd->pd_proc == NULL,
271 ("procdesc_free: pd_proc != NULL"));
272 KASSERT((pd->pd_flags & PDF_CLOSED),
273 ("procdesc_free: !PDF_CLOSED"));
275 PROCDESC_LOCK_DESTROY(pd);
276 uma_zfree(procdesc_zone, pd);
281 * procdesc_exit() - notify a process descriptor that its process is exiting.
282 * We use the proctree_lock to ensure that process exit either happens
283 * strictly before or strictly after a concurrent call to procdesc_close().
286 procdesc_exit(struct proc *p)
290 sx_assert(&proctree_lock, SA_XLOCKED);
291 PROC_LOCK_ASSERT(p, MA_OWNED);
292 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL"));
297 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc,
298 ("procdesc_exit: closed && parent not init"));
300 pd->pd_flags |= PDF_EXITED;
303 * If the process descriptor has been closed, then we have nothing
304 * to do; return 1 so that init will get SIGCHLD and do the reaping.
305 * Clean up the procdesc now rather than letting it happen during
308 if (pd->pd_flags & PDF_CLOSED) {
311 p->p_procdesc = NULL;
315 if (pd->pd_flags & PDF_SELECTED) {
316 pd->pd_flags &= ~PDF_SELECTED;
317 selwakeup(&pd->pd_selinfo);
324 * When a process descriptor is reaped, perhaps as a result of close() or
325 * pdwait4(), release the process's reference on the process descriptor.
328 procdesc_reap(struct proc *p)
332 sx_assert(&proctree_lock, SA_XLOCKED);
333 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
337 p->p_procdesc = NULL;
342 * procdesc_close() - last close on a process descriptor. If the process is
343 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let
344 * init(8) clean up the mess; if not, we have to clean up the zombie ourselves.
347 procdesc_close(struct file *fp, struct thread *td)
352 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc"));
355 fp->f_ops = &badfileops;
358 sx_xlock(&proctree_lock);
360 pd->pd_flags |= PDF_CLOSED;
365 * This is the case where process' exit status was already
366 * collected and procdesc_reap() was already called.
368 sx_xunlock(&proctree_lock);
369 } else if (p->p_state == PRS_ZOMBIE) {
371 * If the process is already dead and just awaiting reaping,
372 * do that now. This will release the process's reference to
373 * the process descriptor when it calls back into
378 proc_reap(curthread, p, NULL, 0);
381 * If the process is not yet dead, we need to kill it, but we
382 * can't wait around synchronously for it to go away, as that
383 * path leads to madness (and deadlocks). First, detach the
384 * process from its descriptor so that its exit status will
385 * be reported normally.
389 p->p_procdesc = NULL;
393 * Next, reparent it to init(8) so that there's someone to
394 * pick up the pieces; finally, terminate with prejudice.
396 p->p_sigparent = SIGCHLD;
397 proc_reparent(p, initproc);
398 if ((pd->pd_flags & PDF_DAEMON) == 0)
399 kern_psignal(p, SIGKILL);
401 sx_xunlock(&proctree_lock);
405 * Release the file descriptor's reference on the process descriptor.
412 procdesc_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
413 int flags, struct thread *td)
420 procdesc_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
421 int flags, struct thread *td)
428 procdesc_truncate(struct file *fp, off_t length, struct ucred *active_cred,
436 procdesc_ioctl(struct file *fp, u_long com, void *data,
437 struct ucred *active_cred, struct thread *td)
444 procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
453 if (pd->pd_flags & PDF_EXITED)
456 selrecord(td, &pd->pd_selinfo);
457 pd->pd_flags |= PDF_SELECTED;
464 procdesc_kqfilter(struct file *fp, struct knote *kn)
471 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
475 struct timeval pstart;
478 * XXXRW: Perhaps we should cache some more information from the
479 * process so that we can return it reliably here even after it has
480 * died. For example, caching its credential data.
482 bzero(sb, sizeof(*sb));
484 sx_slock(&proctree_lock);
485 if (pd->pd_proc != NULL) {
486 PROC_LOCK(pd->pd_proc);
488 /* Set birth and [acm] times to process start time. */
489 pstart = pd->pd_proc->p_stats->p_start;
490 timevaladd(&pstart, &boottime);
491 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim);
492 sb->st_atim = sb->st_birthtim;
493 sb->st_ctim = sb->st_birthtim;
494 sb->st_mtim = sb->st_birthtim;
495 if (pd->pd_proc->p_state != PRS_ZOMBIE)
496 sb->st_mode = S_IFREG | S_IRWXU;
498 sb->st_mode = S_IFREG;
499 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid;
500 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid;
501 PROC_UNLOCK(pd->pd_proc);
503 sb->st_mode = S_IFREG;
504 sx_sunlock(&proctree_lock);
509 procdesc_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
517 procdesc_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
524 #else /* !PROCDESC */
527 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
533 #endif /* PROCDESC */