xref: /freebsd/sys/kern/sys_procdesc.c (revision 9a14aa017b21c292740c00ee098195cd46642730)
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_flags = DFLAG_PASSABLE,
117 };
118 
119 /*
120  * Initialize with VFS so that process descriptors are available along with
121  * other file descriptor types.  As long as it runs before init(8) starts,
122  * there shouldn't be a problem.
123  */
124 static void
125 procdesc_init(void *dummy __unused)
126 {
127 
128 	procdesc_zone = uma_zcreate("procdesc", sizeof(struct procdesc),
129 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
130 	if (procdesc_zone == NULL)
131 		panic("procdesc_init: procdesc_zone not initialized");
132 }
133 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL);
134 
135 /*
136  * Return a locked process given a process descriptor, or ESRCH if it has
137  * died.
138  */
139 int
140 procdesc_find(struct thread *td, int fd, cap_rights_t rights,
141     struct proc **p)
142 {
143 	struct procdesc *pd;
144 	struct file *fp;
145 	int error;
146 
147 	error = fget(td, fd, rights, &fp);
148 	if (error)
149 		return (error);
150 	if (fp->f_type != DTYPE_PROCDESC) {
151 		error = EBADF;
152 		goto out;
153 	}
154 	pd = fp->f_data;
155 	sx_slock(&proctree_lock);
156 	if (pd->pd_proc != NULL) {
157 		*p = pd->pd_proc;
158 		PROC_LOCK(*p);
159 	} else
160 		error = ESRCH;
161 	sx_sunlock(&proctree_lock);
162 out:
163 	fdrop(fp, td);
164 	return (error);
165 }
166 
167 /*
168  * Function to be used by procstat(1) sysctls when returning procdesc
169  * information.
170  */
171 pid_t
172 procdesc_pid(struct file *fp_procdesc)
173 {
174 	struct procdesc *pd;
175 
176 	KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC,
177 	   ("procdesc_pid: !procdesc"));
178 
179 	pd = fp_procdesc->f_data;
180 	return (pd->pd_pid);
181 }
182 
183 /*
184  * Retrieve the PID associated with a process descriptor.
185  */
186 int
187 kern_pdgetpid(struct thread *td, int fd, cap_rights_t rights, pid_t *pidp)
188 {
189 	struct file *fp;
190 	int error;
191 
192 	error = fget(td, fd, rights, &fp);
193 	if (error)
194 		return (error);
195 	if (fp->f_type != DTYPE_PROCDESC) {
196 		error = EBADF;
197 		goto out;
198 	}
199 	*pidp = procdesc_pid(fp);
200 out:
201 	fdrop(fp, td);
202 	return (error);
203 }
204 
205 /*
206  * System call to return the pid of a process given its process descriptor.
207  */
208 int
209 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
210 {
211 	pid_t pid;
212 	int error;
213 
214 	AUDIT_ARG_FD(uap->fd);
215 	error = kern_pdgetpid(td, uap->fd, CAP_PDGETPID, &pid);
216 	if (error == 0)
217 		error = copyout(&pid, uap->pidp, sizeof(pid));
218 	return (error);
219 }
220 
221 /*
222  * When a new process is forked by pdfork(), a file descriptor is allocated
223  * by the fork code first, then the process is forked, and then we get a
224  * chance to set up the process descriptor.  Failure is not permitted at this
225  * point, so procdesc_new() must succeed.
226  */
227 void
228 procdesc_new(struct proc *p, int flags)
229 {
230 	struct procdesc *pd;
231 
232 	pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO);
233 	pd->pd_proc = p;
234 	pd->pd_pid = p->p_pid;
235 	p->p_procdesc = pd;
236 	pd->pd_flags = 0;
237 	if (flags & PD_DAEMON)
238 		pd->pd_flags |= PDF_DAEMON;
239 	PROCDESC_LOCK_INIT(pd);
240 
241 	/*
242 	 * Process descriptors start out with two references: one from their
243 	 * struct file, and the other from their struct proc.
244 	 */
245 	refcount_init(&pd->pd_refcount, 2);
246 }
247 
248 /*
249  * Initialize a file with a process descriptor.
250  */
251 void
252 procdesc_finit(struct procdesc *pdp, struct file *fp)
253 {
254 
255 	finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops);
256 }
257 
258 static void
259 procdesc_free(struct procdesc *pd)
260 {
261 
262 	/*
263 	 * When the last reference is released, we assert that the descriptor
264 	 * has been closed, but not that the process has exited, as we will
265 	 * detach the descriptor before the process dies if the descript is
266 	 * closed, as we can't wait synchronously.
267 	 */
268 	if (refcount_release(&pd->pd_refcount)) {
269 		KASSERT(pd->pd_proc == NULL,
270 		    ("procdesc_free: pd_proc != NULL"));
271 		KASSERT((pd->pd_flags & PDF_CLOSED),
272 		    ("procdesc_free: !PDF_CLOSED"));
273 
274 		PROCDESC_LOCK_DESTROY(pd);
275 		uma_zfree(procdesc_zone, pd);
276 	}
277 }
278 
279 /*
280  * procdesc_exit() - notify a process descriptor that its process is exiting.
281  * We use the proctree_lock to ensure that process exit either happens
282  * strictly before or strictly after a concurrent call to procdesc_close().
283  */
284 int
285 procdesc_exit(struct proc *p)
286 {
287 	struct procdesc *pd;
288 
289 	sx_assert(&proctree_lock, SA_XLOCKED);
290 	PROC_LOCK_ASSERT(p, MA_OWNED);
291 	KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL"));
292 
293 	pd = p->p_procdesc;
294 
295 	PROCDESC_LOCK(pd);
296 	KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc,
297 	    ("procdesc_exit: closed && parent not init"));
298 
299 	pd->pd_flags |= PDF_EXITED;
300 
301 	/*
302 	 * If the process descriptor has been closed, then we have nothing
303 	 * to do; return 1 so that init will get SIGCHLD and do the reaping.
304 	 * Clean up the procdesc now rather than letting it happen during
305 	 * that reap.
306 	 */
307 	if (pd->pd_flags & PDF_CLOSED) {
308 		PROCDESC_UNLOCK(pd);
309 		pd->pd_proc = NULL;
310 		p->p_procdesc = NULL;
311 		procdesc_free(pd);
312 		return (1);
313 	}
314 	if (pd->pd_flags & PDF_SELECTED) {
315 		pd->pd_flags &= ~PDF_SELECTED;
316 		selwakeup(&pd->pd_selinfo);
317 	}
318 	PROCDESC_UNLOCK(pd);
319 	return (0);
320 }
321 
322 /*
323  * When a process descriptor is reaped, perhaps as a result of close() or
324  * pdwait4(), release the process's reference on the process descriptor.
325  */
326 void
327 procdesc_reap(struct proc *p)
328 {
329 	struct procdesc *pd;
330 
331 	sx_assert(&proctree_lock, SA_XLOCKED);
332 	KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL"));
333 
334 	pd = p->p_procdesc;
335 	pd->pd_proc = NULL;
336 	procdesc_free(pd);
337 }
338 
339 /*
340  * procdesc_close() - last close on a process descriptor.  If the process is
341  * still running, terminate with SIGKILL (unless PD_DAEMON is set) and let
342  * init(8) clean up the mess; if not, we have to clean up the zombie ourselves.
343  */
344 static int
345 procdesc_close(struct file *fp, struct thread *td)
346 {
347 	struct procdesc *pd;
348 	struct proc *p;
349 
350 	KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc"));
351 
352 	pd = fp->f_data;
353 	fp->f_ops = &badfileops;
354 	fp->f_data = NULL;
355 
356 	sx_xlock(&proctree_lock);
357 	PROCDESC_LOCK(pd);
358 	pd->pd_flags |= PDF_CLOSED;
359 	PROCDESC_UNLOCK(pd);
360 	p = pd->pd_proc;
361 	PROC_LOCK(p);
362 	if (p->p_state == PRS_ZOMBIE) {
363 		/*
364 		 * If the process is already dead and just awaiting reaping,
365 		 * do that now.  This will release the process's reference to
366 		 * the process descriptor when it calls back into
367 		 * procdesc_reap().
368 		 */
369 		PROC_SLOCK(p);
370 		proc_reap(curthread, p, NULL, 0, NULL);
371 	} else {
372 		/*
373 		 * If the process is not yet dead, we need to kill it, but we
374 		 * can't wait around synchronously for it to go away, as that
375 		 * path leads to madness (and deadlocks).  First, detach the
376 		 * process from its descriptor so that its exit status will
377 		 * be reported normally.
378 		 */
379 		pd->pd_proc = NULL;
380 		p->p_procdesc = NULL;
381 		procdesc_free(pd);
382 
383 		/*
384 		 * Next, reparent it to init(8) so that there's someone to
385 		 * pick up the pieces; finally, terminate with prejudice.
386 		 */
387 		p->p_sigparent = SIGCHLD;
388 		proc_reparent(p, initproc);
389 		if ((pd->pd_flags & PD_DAEMON) == 0)
390 			kern_psignal(p, SIGKILL);
391 		PROC_UNLOCK(p);
392 		sx_xunlock(&proctree_lock);
393 	}
394 
395 	/*
396 	 * Release the file descriptor's reference on the process descriptor.
397 	 */
398 	procdesc_free(pd);
399 	return (0);
400 }
401 
402 static int
403 procdesc_read(struct file *fp, struct uio *uio, struct ucred *active_cred,
404     int flags, struct thread *td)
405 {
406 
407 	return (EOPNOTSUPP);
408 }
409 
410 static int
411 procdesc_write(struct file *fp, struct uio *uio, struct ucred *active_cred,
412     int flags, struct thread *td)
413 {
414 
415 	return (EOPNOTSUPP);
416 }
417 
418 static int
419 procdesc_truncate(struct file *fp, off_t length, struct ucred *active_cred,
420     struct thread *td)
421 {
422 
423 	return (EOPNOTSUPP);
424 }
425 
426 static int
427 procdesc_ioctl(struct file *fp, u_long com, void *data,
428     struct ucred *active_cred, struct thread *td)
429 {
430 
431 	return (EOPNOTSUPP);
432 }
433 
434 static int
435 procdesc_poll(struct file *fp, int events, struct ucred *active_cred,
436     struct thread *td)
437 {
438 	struct procdesc *pd;
439 	int revents;
440 
441 	revents = 0;
442 	pd = fp->f_data;
443 	PROCDESC_LOCK(pd);
444 	if (pd->pd_flags & PDF_EXITED)
445 		revents |= POLLHUP;
446 	if (revents == 0) {
447 		selrecord(td, &pd->pd_selinfo);
448 		pd->pd_flags |= PDF_SELECTED;
449 	}
450 	PROCDESC_UNLOCK(pd);
451 	return (revents);
452 }
453 
454 static int
455 procdesc_kqfilter(struct file *fp, struct knote *kn)
456 {
457 
458 	return (EOPNOTSUPP);
459 }
460 
461 static int
462 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
463     struct thread *td)
464 {
465 	struct procdesc *pd;
466 	struct timeval pstart;
467 
468 	/*
469 	 * XXXRW: Perhaps we should cache some more information from the
470 	 * process so that we can return it reliably here even after it has
471 	 * died.  For example, caching its credential data.
472 	 */
473 	bzero(sb, sizeof(*sb));
474 	pd = fp->f_data;
475 	sx_slock(&proctree_lock);
476 	if (pd->pd_proc != NULL) {
477 		PROC_LOCK(pd->pd_proc);
478 
479 		/* Set birth and [acm] times to process start time. */
480 		pstart = pd->pd_proc->p_stats->p_start;
481 		timevaladd(&pstart, &boottime);
482 		TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim);
483 		sb->st_atim = sb->st_birthtim;
484 		sb->st_ctim = sb->st_birthtim;
485 		sb->st_mtim = sb->st_birthtim;
486 		if (pd->pd_proc->p_state != PRS_ZOMBIE)
487 			sb->st_mode = S_IFREG | S_IRWXU;
488 		else
489 			sb->st_mode = S_IFREG;
490 		sb->st_uid = pd->pd_proc->p_ucred->cr_ruid;
491 		sb->st_gid = pd->pd_proc->p_ucred->cr_rgid;
492 		PROC_UNLOCK(pd->pd_proc);
493 	} else
494 		sb->st_mode = S_IFREG;
495 	sx_sunlock(&proctree_lock);
496 	return (0);
497 }
498 
499 static int
500 procdesc_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
501     struct thread *td)
502 {
503 
504 	return (EOPNOTSUPP);
505 }
506 
507 static int
508 procdesc_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
509     struct thread *td)
510 {
511 
512 	return (EOPNOTSUPP);
513 }
514 
515 #else /* !PROCDESC */
516 
517 int
518 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap)
519 {
520 
521 	return (ENOSYS);
522 }
523 
524 #endif /* PROCDESC */
525