xref: /freebsd/sys/kern/sys_procdesc.c (revision a18eacbefdfa1085ca3db829e86ece78cd416493)
1 /*-
2  * Copyright (c) 2009 Robert N. M. Watson
3  * All rights reserved.
4  *
5  * This software was developed at the University of Cambridge Computer
6  * Laboratory with support from a grant from Google, Inc.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
16  *
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
27  * SUCH DAMAGE.
28  */
29 
30 /*-
31  * FreeBSD process descriptor facility.
32  *
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).
38  *
39  * Some semantics:
40  *
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.
53  *
54  * Open questions:
55  *
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)?
59  */
60 
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
63 
64 #include "opt_procdesc.h"
65 
66 #include <sys/param.h>
67 #include <sys/capability.h>
68 #include <sys/fcntl.h>
69 #include <sys/file.h>
70 #include <sys/filedesc.h>
71 #include <sys/kernel.h>
72 #include <sys/lock.h>
73 #include <sys/mutex.h>
74 #include <sys/poll.h>
75 #include <sys/proc.h>
76 #include <sys/procdesc.h>
77 #include <sys/resourcevar.h>
78 #include <sys/stat.h>
79 #include <sys/sysproto.h>
80 #include <sys/sysctl.h>
81 #include <sys/systm.h>
82 #include <sys/ucred.h>
83 
84 #include <security/audit/audit.h>
85 
86 #include <vm/uma.h>
87 
88 #ifdef PROCDESC
89 
90 FEATURE(process_descriptors, "Process Descriptors");
91 
92 static uma_zone_t procdesc_zone;
93 
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;
104 
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,
118 };
119 
120 /*
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.
124  */
125 static void
126 procdesc_init(void *dummy __unused)
127 {
128 
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");
133 }
134 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL);
135 
136 /*
137  * Return a locked process given a process descriptor, or ESRCH if it has
138  * died.
139  */
140 int
141 procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp,
142     struct proc **p)
143 {
144 	struct procdesc *pd;
145 	struct file *fp;
146 	int error;
147 
148 	error = fget(td, fd, rightsp, &fp);
149 	if (error)
150 		return (error);
151 	if (fp->f_type != DTYPE_PROCDESC) {
152 		error = EBADF;
153 		goto out;
154 	}
155 	pd = fp->f_data;
156 	sx_slock(&proctree_lock);
157 	if (pd->pd_proc != NULL) {
158 		*p = pd->pd_proc;
159 		PROC_LOCK(*p);
160 	} else
161 		error = ESRCH;
162 	sx_sunlock(&proctree_lock);
163 out:
164 	fdrop(fp, td);
165 	return (error);
166 }
167 
168 /*
169  * Function to be used by procstat(1) sysctls when returning procdesc
170  * information.
171  */
172 pid_t
173 procdesc_pid(struct file *fp_procdesc)
174 {
175 	struct procdesc *pd;
176 
177 	KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC,
178 	   ("procdesc_pid: !procdesc"));
179 
180 	pd = fp_procdesc->f_data;
181 	return (pd->pd_pid);
182 }
183 
184 /*
185  * Retrieve the PID associated with a process descriptor.
186  */
187 int
188 kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp)
189 {
190 	struct file *fp;
191 	int error;
192 
193 	error = fget(td, fd, rightsp, &fp);
194 	if (error)
195 		return (error);
196 	if (fp->f_type != DTYPE_PROCDESC) {
197 		error = EBADF;
198 		goto out;
199 	}
200 	*pidp = procdesc_pid(fp);
201 out:
202 	fdrop(fp, td);
203 	return (error);
204 }
205 
206 /*
207  * System call to return the pid of a process given its process descriptor.
208  */
209 int
210 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
211 {
212 	cap_rights_t rights;
213 	pid_t pid;
214 	int error;
215 
216 	AUDIT_ARG_FD(uap->fd);
217 	error = kern_pdgetpid(td, uap->fd,
218 	    cap_rights_init(&rights, CAP_PDGETPID), &pid);
219 	if (error == 0)
220 		error = copyout(&pid, uap->pidp, sizeof(pid));
221 	return (error);
222 }
223 
224 /*
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.
229  */
230 void
231 procdesc_new(struct proc *p, int flags)
232 {
233 	struct procdesc *pd;
234 
235 	pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO);
236 	pd->pd_proc = p;
237 	pd->pd_pid = p->p_pid;
238 	p->p_procdesc = pd;
239 	pd->pd_flags = 0;
240 	if (flags & PD_DAEMON)
241 		pd->pd_flags |= PDF_DAEMON;
242 	PROCDESC_LOCK_INIT(pd);
243 
244 	/*
245 	 * Process descriptors start out with two references: one from their
246 	 * struct file, and the other from their struct proc.
247 	 */
248 	refcount_init(&pd->pd_refcount, 2);
249 }
250 
251 /*
252  * Initialize a file with a process descriptor.
253  */
254 void
255 procdesc_finit(struct procdesc *pdp, struct file *fp)
256 {
257 
258 	finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
259 }
260 
261 static void
262 procdesc_free(struct procdesc *pd)
263 {
264 
265 	/*
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.
270 	 */
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"));
276 
277 		PROCDESC_LOCK_DESTROY(pd);
278 		uma_zfree(procdesc_zone, pd);
279 	}
280 }
281 
282 /*
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().
286  */
287 int
288 procdesc_exit(struct proc *p)
289 {
290 	struct procdesc *pd;
291 
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"));
295 
296 	pd = p->p_procdesc;
297 
298 	PROCDESC_LOCK(pd);
299 	KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc,
300 	    ("procdesc_exit: closed && parent not init"));
301 
302 	pd->pd_flags |= PDF_EXITED;
303 
304 	/*
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
308 	 * that reap.
309 	 */
310 	if (pd->pd_flags & PDF_CLOSED) {
311 		PROCDESC_UNLOCK(pd);
312 		pd->pd_proc = NULL;
313 		p->p_procdesc = NULL;
314 		procdesc_free(pd);
315 		return (1);
316 	}
317 	if (pd->pd_flags & PDF_SELECTED) {
318 		pd->pd_flags &= ~PDF_SELECTED;
319 		selwakeup(&pd->pd_selinfo);
320 	}
321 	PROCDESC_UNLOCK(pd);
322 	return (0);
323 }
324 
325 /*
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.
328  */
329 void
330 procdesc_reap(struct proc *p)
331 {
332 	struct procdesc *pd;
333 
334 	sx_assert(&proctree_lock, SA_XLOCKED);
335 	KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
336 
337 	pd = p->p_procdesc;
338 	pd->pd_proc = NULL;
339 	p->p_procdesc = NULL;
340 	procdesc_free(pd);
341 }
342 
343 /*
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.
347  */
348 static int
349 procdesc_close(struct file *fp, struct thread *td)
350 {
351 	struct procdesc *pd;
352 	struct proc *p;
353 
354 	KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc"));
355 
356 	pd = fp->f_data;
357 	fp->f_ops = &badfileops;
358 	fp->f_data = NULL;
359 
360 	sx_xlock(&proctree_lock);
361 	PROCDESC_LOCK(pd);
362 	pd->pd_flags |= PDF_CLOSED;
363 	PROCDESC_UNLOCK(pd);
364 	p = pd->pd_proc;
365 	if (p == NULL) {
366 		/*
367 		 * This is the case where process' exit status was already
368 		 * collected and procdesc_reap() was already called.
369 		 */
370 		sx_xunlock(&proctree_lock);
371 	} else if (p->p_state == PRS_ZOMBIE) {
372 		/*
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
376 		 * procdesc_reap().
377 		 */
378 		PROC_LOCK(p);
379 		PROC_SLOCK(p);
380 		proc_reap(curthread, p, NULL, 0);
381 	} else {
382 		/*
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.
388 		 */
389 		PROC_LOCK(p);
390 		pd->pd_proc = NULL;
391 		p->p_procdesc = NULL;
392 		procdesc_free(pd);
393 
394 		/*
395 		 * Next, reparent it to init(8) so that there's someone to
396 		 * pick up the pieces; finally, terminate with prejudice.
397 		 */
398 		p->p_sigparent = SIGCHLD;
399 		proc_reparent(p, initproc);
400 		if ((pd->pd_flags & PDF_DAEMON) == 0)
401 			kern_psignal(p, SIGKILL);
402 		PROC_UNLOCK(p);
403 		sx_xunlock(&proctree_lock);
404 	}
405 
406 	/*
407 	 * Release the file descriptor's reference on the process descriptor.
408 	 */
409 	procdesc_free(pd);
410 	return (0);
411 }
412 
413 static int
414 procdesc_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
415     int flags, struct thread *td)
416 {
417 
418 	return (EOPNOTSUPP);
419 }
420 
421 static int
422 procdesc_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
423     int flags, struct thread *td)
424 {
425 
426 	return (EOPNOTSUPP);
427 }
428 
429 static int
430 procdesc_truncate(struct file *fp, off_t length, struct ucred *active_cred,
431     struct thread *td)
432 {
433 
434 	return (EOPNOTSUPP);
435 }
436 
437 static int
438 procdesc_ioctl(struct file *fp, u_long com, void *data,
439     struct ucred *active_cred, struct thread *td)
440 {
441 
442 	return (EOPNOTSUPP);
443 }
444 
445 static int
446 procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
447     struct thread *td)
448 {
449 	struct procdesc *pd;
450 	int revents;
451 
452 	revents = 0;
453 	pd = fp->f_data;
454 	PROCDESC_LOCK(pd);
455 	if (pd->pd_flags & PDF_EXITED)
456 		revents |= POLLHUP;
457 	if (revents == 0) {
458 		selrecord(td, &pd->pd_selinfo);
459 		pd->pd_flags |= PDF_SELECTED;
460 	}
461 	PROCDESC_UNLOCK(pd);
462 	return (revents);
463 }
464 
465 static int
466 procdesc_kqfilter(struct file *fp, struct knote *kn)
467 {
468 
469 	return (EOPNOTSUPP);
470 }
471 
472 static int
473 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
474     struct thread *td)
475 {
476 	struct procdesc *pd;
477 	struct timeval pstart;
478 
479 	/*
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.
483 	 */
484 	bzero(sb, sizeof(*sb));
485 	pd = fp->f_data;
486 	sx_slock(&proctree_lock);
487 	if (pd->pd_proc != NULL) {
488 		PROC_LOCK(pd->pd_proc);
489 
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;
499 		else
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);
504 	} else
505 		sb->st_mode = S_IFREG;
506 	sx_sunlock(&proctree_lock);
507 	return (0);
508 }
509 
510 static int
511 procdesc_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
512     struct thread *td)
513 {
514 
515 	return (EOPNOTSUPP);
516 }
517 
518 static int
519 procdesc_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
520     struct thread *td)
521 {
522 
523 	return (EOPNOTSUPP);
524 }
525 
526 #else /* !PROCDESC */
527 
528 int
529 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
530 {
531 
532 	return (ENOSYS);
533 }
534 
535 #endif /* PROCDESC */
536