xref: /illumos-gate/usr/src/uts/common/fs/proc/prvnops.c (revision dd72704bd9e794056c558153663c739e2012d721)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2018, Joyent, Inc.
25  * Copyright (c) 2017 by Delphix. All rights reserved.
26  * Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
27  * Copyright 2022 MNX Cloud, Inc.
28  * Copyright 2022 Oxide Computer Company
29  */
30 
31 /*	Copyright (c) 1984,	 1986, 1987, 1988, 1989 AT&T	*/
32 /*	  All Rights Reserved	*/
33 
34 #include <sys/types.h>
35 #include <sys/param.h>
36 #include <sys/time.h>
37 #include <sys/cred.h>
38 #include <sys/policy.h>
39 #include <sys/debug.h>
40 #include <sys/dirent.h>
41 #include <sys/errno.h>
42 #include <sys/file.h>
43 #include <sys/inline.h>
44 #include <sys/kmem.h>
45 #include <sys/pathname.h>
46 #include <sys/proc.h>
47 #include <sys/brand.h>
48 #include <sys/signal.h>
49 #include <sys/stat.h>
50 #include <sys/sysmacros.h>
51 #include <sys/systm.h>
52 #include <sys/zone.h>
53 #include <sys/uio.h>
54 #include <sys/var.h>
55 #include <sys/mode.h>
56 #include <sys/poll.h>
57 #include <sys/user.h>
58 #include <sys/vfs.h>
59 #include <sys/vfs_opreg.h>
60 #include <sys/gfs.h>
61 #include <sys/vnode.h>
62 #include <sys/fault.h>
63 #include <sys/syscall.h>
64 #include <sys/procfs.h>
65 #include <sys/atomic.h>
66 #include <sys/cmn_err.h>
67 #include <sys/contract_impl.h>
68 #include <sys/ctfs.h>
69 #include <sys/avl.h>
70 #include <sys/ctype.h>
71 #include <fs/fs_subr.h>
72 #include <vm/rm.h>
73 #include <vm/as.h>
74 #include <vm/seg.h>
75 #include <vm/seg_vn.h>
76 #include <vm/hat.h>
77 #include <fs/proc/prdata.h>
78 #if defined(__sparc)
79 #include <sys/regset.h>
80 #endif
81 #if defined(__x86)
82 #include <sys/sysi86.h>
83 #endif
84 
85 /*
86  * Created by prinit.
87  */
88 vnodeops_t *prvnodeops;
89 
90 /*
91  * Directory characteristics (patterned after the s5 file system).
92  */
93 #define	PRROOTINO	2
94 
95 #define	PRDIRSIZE	14
96 struct prdirect {
97 	ushort_t	d_ino;
98 	char		d_name[PRDIRSIZE];
99 };
100 
101 #define	PRSDSIZE	(sizeof (struct prdirect))
102 
103 /*
104  * Directory characteristics.
105  */
106 typedef struct prdirent {
107 	ino64_t		d_ino;		/* "inode number" of entry */
108 	off64_t		d_off;		/* offset of disk directory entry */
109 	unsigned short	d_reclen;	/* length of this record */
110 	char		d_name[14];	/* name of file */
111 } prdirent_t;
112 
113 /*
114  * Contents of a /proc/<pid> directory.
115  * Reuse d_ino field for the /proc file type.
116  */
117 static prdirent_t piddir[] = {
118 	{ PR_PIDDIR,	 1 * sizeof (prdirent_t), sizeof (prdirent_t),
119 		"." },
120 	{ PR_PROCDIR,	 2 * sizeof (prdirent_t), sizeof (prdirent_t),
121 		".." },
122 	{ PR_AS,	 3 * sizeof (prdirent_t), sizeof (prdirent_t),
123 		"as" },
124 	{ PR_CTL,	 4 * sizeof (prdirent_t), sizeof (prdirent_t),
125 		"ctl" },
126 	{ PR_STATUS,	 5 * sizeof (prdirent_t), sizeof (prdirent_t),
127 		"status" },
128 	{ PR_LSTATUS,	 6 * sizeof (prdirent_t), sizeof (prdirent_t),
129 		"lstatus" },
130 	{ PR_PSINFO,	 7 * sizeof (prdirent_t), sizeof (prdirent_t),
131 		"psinfo" },
132 	{ PR_LPSINFO,	 8 * sizeof (prdirent_t), sizeof (prdirent_t),
133 		"lpsinfo" },
134 	{ PR_MAP,	 9 * sizeof (prdirent_t), sizeof (prdirent_t),
135 		"map" },
136 	{ PR_RMAP,	10 * sizeof (prdirent_t), sizeof (prdirent_t),
137 		"rmap" },
138 	{ PR_XMAP,	11 * sizeof (prdirent_t), sizeof (prdirent_t),
139 		"xmap" },
140 	{ PR_CRED,	12 * sizeof (prdirent_t), sizeof (prdirent_t),
141 		"cred" },
142 	{ PR_SIGACT,	13 * sizeof (prdirent_t), sizeof (prdirent_t),
143 		"sigact" },
144 	{ PR_AUXV,	14 * sizeof (prdirent_t), sizeof (prdirent_t),
145 		"auxv" },
146 	{ PR_USAGE,	15 * sizeof (prdirent_t), sizeof (prdirent_t),
147 		"usage" },
148 	{ PR_LUSAGE,	16 * sizeof (prdirent_t), sizeof (prdirent_t),
149 		"lusage" },
150 	{ PR_PAGEDATA,	17 * sizeof (prdirent_t), sizeof (prdirent_t),
151 		"pagedata" },
152 	{ PR_WATCH,	18 * sizeof (prdirent_t), sizeof (prdirent_t),
153 		"watch" },
154 	{ PR_CURDIR,	19 * sizeof (prdirent_t), sizeof (prdirent_t),
155 		"cwd" },
156 	{ PR_ROOTDIR,	20 * sizeof (prdirent_t), sizeof (prdirent_t),
157 		"root" },
158 	{ PR_FDDIR,	21 * sizeof (prdirent_t), sizeof (prdirent_t),
159 		"fd" },
160 	{ PR_FDINFODIR,	22 * sizeof (prdirent_t), sizeof (prdirent_t),
161 		"fdinfo" },
162 	{ PR_OBJECTDIR,	23 * sizeof (prdirent_t), sizeof (prdirent_t),
163 		"object" },
164 	{ PR_LWPDIR,	24 * sizeof (prdirent_t), sizeof (prdirent_t),
165 		"lwp" },
166 	{ PR_PRIV,	25 * sizeof (prdirent_t), sizeof (prdirent_t),
167 		"priv" },
168 	{ PR_PATHDIR,	26 * sizeof (prdirent_t), sizeof (prdirent_t),
169 		"path" },
170 	{ PR_CTDIR,	27 * sizeof (prdirent_t), sizeof (prdirent_t),
171 		"contracts" },
172 	{ PR_SECFLAGS,	28 * sizeof (prdirent_t), sizeof (prdirent_t),
173 		"secflags" },
174 #if defined(__x86)
175 	{ PR_LDT,	29 * sizeof (prdirent_t), sizeof (prdirent_t),
176 		"ldt" },
177 #endif
178 };
179 
180 #define	NPIDDIRFILES	(sizeof (piddir) / sizeof (piddir[0]) - 2)
181 
182 /*
183  * Contents of a /proc/<pid>/lwp/<lwpid> directory.
184  */
185 static prdirent_t lwpiddir[] = {
186 	{ PR_LWPIDDIR,	 1 * sizeof (prdirent_t), sizeof (prdirent_t),
187 		"." },
188 	{ PR_LWPDIR,	 2 * sizeof (prdirent_t), sizeof (prdirent_t),
189 		".." },
190 	{ PR_LWPCTL,	 3 * sizeof (prdirent_t), sizeof (prdirent_t),
191 		"lwpctl" },
192 	{ PR_LWPNAME,	 4 * sizeof (prdirent_t), sizeof (prdirent_t),
193 		"lwpname" },
194 	{ PR_LWPSTATUS,	 5 * sizeof (prdirent_t), sizeof (prdirent_t),
195 		"lwpstatus" },
196 	{ PR_LWPSINFO,	 6 * sizeof (prdirent_t), sizeof (prdirent_t),
197 		"lwpsinfo" },
198 	{ PR_LWPUSAGE,	 7 * sizeof (prdirent_t), sizeof (prdirent_t),
199 		"lwpusage" },
200 	{ PR_XREGS,	 8 * sizeof (prdirent_t), sizeof (prdirent_t),
201 		"xregs" },
202 	{ PR_TMPLDIR,	 9 * sizeof (prdirent_t), sizeof (prdirent_t),
203 		"templates" },
204 	{ PR_SPYMASTER,	 10 * sizeof (prdirent_t), sizeof (prdirent_t),
205 		"spymaster" },
206 #if defined(__sparc)
207 	{ PR_GWINDOWS,	11 * sizeof (prdirent_t), sizeof (prdirent_t),
208 		"gwindows" },
209 	{ PR_ASRS,	12 * sizeof (prdirent_t), sizeof (prdirent_t),
210 		"asrs" },
211 #endif
212 };
213 
214 #define	NLWPIDDIRFILES	(sizeof (lwpiddir) / sizeof (lwpiddir[0]) - 2)
215 
216 /*
217  * Span of entries in the array files (lstatus, lpsinfo, lusage).
218  * We make the span larger than the size of the structure on purpose,
219  * to make sure that programs cannot use the structure size by mistake.
220  * Align _ILP32 structures at 8 bytes, _LP64 structures at 16 bytes.
221  */
222 #ifdef _LP64
223 #define	LSPAN(type)	(round16(sizeof (type)) + 16)
224 #define	LSPAN32(type)	(round8(sizeof (type)) + 8)
225 #else
226 #define	LSPAN(type)	(round8(sizeof (type)) + 8)
227 #endif
228 
229 static void rebuild_objdir(struct as *);
230 static void prfreecommon(prcommon_t *);
231 static int praccess(vnode_t *, int, int, cred_t *, caller_context_t *);
232 
233 static int
234 propen(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct)
235 {
236 	vnode_t *vp = *vpp;
237 	prnode_t *pnp = VTOP(vp);
238 	prcommon_t *pcp = pnp->pr_pcommon;
239 	prnodetype_t type = pnp->pr_type;
240 	vnode_t *rvp;
241 	vtype_t vtype;
242 	proc_t *p;
243 	int error = 0;
244 	prnode_t *npnp = NULL;
245 
246 	/*
247 	 * Nothing to do for the /proc directory itself.
248 	 */
249 	if (type == PR_PROCDIR)
250 		return (0);
251 
252 	/*
253 	 * If we are opening an underlying mapped object, reject opens
254 	 * for writing regardless of the objects's access modes.
255 	 * If we are opening a file in the /proc/pid/fd directory,
256 	 * reject the open for any but a regular file or directory.
257 	 * Just do it if we are opening the current or root directory.
258 	 */
259 	switch (type) {
260 	case PR_OBJECT:
261 	case PR_FD:
262 	case PR_CURDIR:
263 	case PR_ROOTDIR:
264 		rvp = pnp->pr_realvp;
265 		vtype = rvp->v_type;
266 		if ((type == PR_OBJECT && (flag & FWRITE)) ||
267 		    (type == PR_FD && vtype != VREG && vtype != VDIR))
268 			error = EACCES;
269 		else {
270 			/*
271 			 * Need to hold rvp since VOP_OPEN() may release it.
272 			 */
273 			VN_HOLD(rvp);
274 			error = VOP_OPEN(&rvp, flag, cr, ct);
275 			if (error) {
276 				VN_RELE(rvp);
277 			} else {
278 				*vpp = rvp;
279 				VN_RELE(vp);
280 			}
281 		}
282 		return (error);
283 	default:
284 		break;
285 	}
286 
287 	/*
288 	 * If we are opening the pagedata file, allocate a prnode now
289 	 * to avoid calling kmem_alloc() while holding p->p_lock.
290 	 */
291 	if (type == PR_PAGEDATA || type == PR_OPAGEDATA)
292 		npnp = prgetnode(vp, type);
293 
294 	/*
295 	 * If the process exists, lock it now.
296 	 * Otherwise we have a race condition with prclose().
297 	 */
298 	p = pr_p_lock(pnp);
299 	mutex_exit(&pr_pidlock);
300 	if (p == NULL) {
301 		if (npnp != NULL)
302 			prfreenode(npnp);
303 		return (ENOENT);
304 	}
305 	ASSERT(p == pcp->prc_proc);
306 	ASSERT(p->p_proc_flag & P_PR_LOCK);
307 
308 	/*
309 	 * Maintain a count of opens for write.  Allow exactly one
310 	 * O_WRITE|O_EXCL request and fail subsequent ones.
311 	 * Don't fail opens of old (bletch!) /proc lwp files.
312 	 * Special case for open by the process itself:
313 	 * Always allow the open by self and discount this
314 	 * open for other opens for writing.
315 	 */
316 	if (flag & FWRITE) {
317 		if (p == curproc) {
318 			pcp->prc_selfopens++;
319 			pnp->pr_flags |= PR_ISSELF;
320 		} else if (type == PR_LWPIDFILE) {
321 			/* EMPTY */;
322 		} else if (flag & FEXCL) {
323 			if (pcp->prc_writers > pcp->prc_selfopens) {
324 				error = EBUSY;
325 				goto out;
326 			}
327 			/* semantic for old /proc interface */
328 			if (type == PR_PIDDIR)
329 				pcp->prc_flags |= PRC_EXCL;
330 		} else if (pcp->prc_flags & PRC_EXCL) {
331 			ASSERT(pcp->prc_writers > pcp->prc_selfopens);
332 			error = secpolicy_proc_excl_open(cr);
333 			if (error)
334 				goto out;
335 		}
336 		pcp->prc_writers++;
337 		/*
338 		 * The vnode may have become invalid between the
339 		 * VOP_LOOKUP() of the /proc vnode and the VOP_OPEN().
340 		 * If so, do now what prinvalidate() should have done.
341 		 */
342 		if ((pnp->pr_flags & PR_INVAL) ||
343 		    (type == PR_PIDDIR &&
344 		    (VTOP(pnp->pr_pidfile)->pr_flags & PR_INVAL))) {
345 			if (p != curproc)
346 				pcp->prc_selfopens++;
347 			ASSERT(pcp->prc_selfopens <= pcp->prc_writers);
348 			if (pcp->prc_selfopens == pcp->prc_writers)
349 				pcp->prc_flags &= ~PRC_EXCL;
350 		}
351 	}
352 
353 	/*
354 	 * If this is a large file open, indicate that in our flags -- some
355 	 * procfs structures are not off_t-neutral (e.g., priovec_t), and
356 	 * the open will need to be differentiated where 32-bit processes
357 	 * pass these structures across the user/kernel boundary.
358 	 */
359 	if (flag & FOFFMAX)
360 		pnp->pr_flags |= PR_OFFMAX;
361 
362 	/*
363 	 * Do file-specific things.
364 	 */
365 	switch (type) {
366 	default:
367 		break;
368 	case PR_PAGEDATA:
369 	case PR_OPAGEDATA:
370 		/*
371 		 * Enable data collection for page data file;
372 		 * get unique id from the hat layer.
373 		 */
374 		{
375 			int id;
376 
377 			/*
378 			 * Drop p->p_lock to call hat_startstat()
379 			 */
380 			mutex_exit(&p->p_lock);
381 			if ((p->p_flag & SSYS) || p->p_as == &kas ||
382 			    (id = hat_startstat(p->p_as)) == -1) {
383 				mutex_enter(&p->p_lock);
384 				error = ENOMEM;
385 			} else if (pnp->pr_hatid == 0) {
386 				mutex_enter(&p->p_lock);
387 				pnp->pr_hatid = (uint_t)id;
388 			} else {
389 				mutex_enter(&p->p_lock);
390 				/*
391 				 * Use our newly allocated prnode.
392 				 */
393 				npnp->pr_hatid = (uint_t)id;
394 				/*
395 				 * prgetnode() initialized most of the prnode.
396 				 * Duplicate the remainder.
397 				 */
398 				npnp->pr_ino = pnp->pr_ino;
399 				npnp->pr_common = pnp->pr_common;
400 				npnp->pr_pcommon = pnp->pr_pcommon;
401 				npnp->pr_parent = pnp->pr_parent;
402 				VN_HOLD(npnp->pr_parent);
403 				npnp->pr_index = pnp->pr_index;
404 
405 				npnp->pr_next = p->p_plist;
406 				p->p_plist = PTOV(npnp);
407 
408 				VN_RELE(PTOV(pnp));
409 				pnp = npnp;
410 				npnp = NULL;
411 				*vpp = PTOV(pnp);
412 			}
413 		}
414 		break;
415 	}
416 
417 out:
418 	prunlock(pnp);
419 
420 	if (npnp != NULL)
421 		prfreenode(npnp);
422 	return (error);
423 }
424 
425 /* ARGSUSED */
426 static int
427 prclose(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr,
428     caller_context_t *ct)
429 {
430 	prnode_t *pnp = VTOP(vp);
431 	prcommon_t *pcp = pnp->pr_pcommon;
432 	prnodetype_t type = pnp->pr_type;
433 	proc_t *p;
434 	kthread_t *t;
435 	user_t *up;
436 
437 	/*
438 	 * Nothing to do for the /proc directory itself.
439 	 */
440 	if (type == PR_PROCDIR)
441 		return (0);
442 
443 	ASSERT(type != PR_OBJECT && type != PR_FD &&
444 	    type != PR_CURDIR && type != PR_ROOTDIR);
445 
446 	/*
447 	 * If the process exists, lock it now.
448 	 * Otherwise we have a race condition with propen().
449 	 * Hold pr_pidlock across the reference to prc_selfopens,
450 	 * and prc_writers in case there is no process anymore,
451 	 * to cover the case of concurrent calls to prclose()
452 	 * after the process has been reaped by freeproc().
453 	 */
454 	p = pr_p_lock(pnp);
455 
456 	/*
457 	 * There is nothing more to do until the last close of
458 	 * the file table entry except to clear the pr_owner
459 	 * field of the prnode and notify any waiters
460 	 * (their file descriptor may have just been closed).
461 	 */
462 	if (count > 1) {
463 		mutex_exit(&pr_pidlock);
464 		if (pnp->pr_owner == curproc && !fisopen(vp))
465 			pnp->pr_owner = NULL;
466 		if (p != NULL) {
467 			prnotify(vp);
468 			prunlock(pnp);
469 		}
470 		return (0);
471 	}
472 
473 	/*
474 	 * Decrement the count of self-opens for writing.
475 	 * Decrement the total count of opens for writing.
476 	 * Cancel exclusive opens when only self-opens remain.
477 	 */
478 	if (flag & FWRITE) {
479 		/*
480 		 * prc_selfopens also contains the count of
481 		 * invalid writers.  See prinvalidate().
482 		 */
483 		if ((pnp->pr_flags & (PR_ISSELF|PR_INVAL)) ||
484 		    (type == PR_PIDDIR &&
485 		    (VTOP(pnp->pr_pidfile)->pr_flags & PR_INVAL))) {
486 			ASSERT(pcp->prc_selfopens != 0);
487 			--pcp->prc_selfopens;
488 		}
489 		ASSERT(pcp->prc_writers != 0);
490 		if (--pcp->prc_writers == pcp->prc_selfopens)
491 			pcp->prc_flags &= ~PRC_EXCL;
492 	}
493 	ASSERT(pcp->prc_writers >= pcp->prc_selfopens);
494 	mutex_exit(&pr_pidlock);
495 	if (pnp->pr_owner == curproc && !fisopen(vp))
496 		pnp->pr_owner = NULL;
497 
498 	/*
499 	 * If there is no process, there is nothing more to do.
500 	 */
501 	if (p == NULL)
502 		return (0);
503 
504 	ASSERT(p == pcp->prc_proc);
505 	prnotify(vp);	/* notify waiters */
506 
507 	/*
508 	 * Do file-specific things.
509 	 */
510 	switch (type) {
511 	default:
512 		break;
513 	case PR_PAGEDATA:
514 	case PR_OPAGEDATA:
515 		/*
516 		 * This is a page data file.
517 		 * Free the hat level statistics.
518 		 * Drop p->p_lock before calling hat_freestat().
519 		 */
520 		mutex_exit(&p->p_lock);
521 		if (p->p_as != &kas && pnp->pr_hatid != 0)
522 			hat_freestat(p->p_as, pnp->pr_hatid);
523 		mutex_enter(&p->p_lock);
524 		pnp->pr_hatid = 0;
525 		break;
526 	}
527 
528 	/*
529 	 * On last close of all writable file descriptors,
530 	 * perform run-on-last-close and/or kill-on-last-close logic.
531 	 * Can't do this is the /proc agent lwp still exists.
532 	 */
533 	if (pcp->prc_writers == 0 &&
534 	    p->p_agenttp == NULL &&
535 	    !(pcp->prc_flags & PRC_DESTROY) &&
536 	    p->p_stat != SZOMB &&
537 	    (p->p_proc_flag & (P_PR_RUNLCL|P_PR_KILLCL))) {
538 		int killproc;
539 
540 		/*
541 		 * Cancel any watchpoints currently in effect.
542 		 * The process might disappear during this operation.
543 		 */
544 		if (pr_cancel_watch(pnp) == NULL)
545 			return (0);
546 		/*
547 		 * If any tracing flags are set, clear them.
548 		 */
549 		if (p->p_proc_flag & P_PR_TRACE) {
550 			up = PTOU(p);
551 			premptyset(&up->u_entrymask);
552 			premptyset(&up->u_exitmask);
553 			up->u_systrap = 0;
554 		}
555 		premptyset(&p->p_sigmask);
556 		premptyset(&p->p_fltmask);
557 		killproc = (p->p_proc_flag & P_PR_KILLCL);
558 		p->p_proc_flag &= ~(P_PR_RUNLCL|P_PR_KILLCL|P_PR_TRACE);
559 		/*
560 		 * Cancel any outstanding single-step requests.
561 		 */
562 		if ((t = p->p_tlist) != NULL) {
563 			/*
564 			 * Drop p_lock because prnostep() touches the stack.
565 			 * The loop is safe because the process is P_PR_LOCK'd.
566 			 */
567 			mutex_exit(&p->p_lock);
568 			do {
569 				prnostep(ttolwp(t));
570 			} while ((t = t->t_forw) != p->p_tlist);
571 			mutex_enter(&p->p_lock);
572 		}
573 		/*
574 		 * Set runnable all lwps stopped by /proc.
575 		 */
576 		if (killproc)
577 			sigtoproc(p, NULL, SIGKILL);
578 		else
579 			allsetrun(p);
580 	}
581 
582 	prunlock(pnp);
583 	return (0);
584 }
585 
586 /*
587  * Array of read functions, indexed by /proc file type.
588  */
589 static int pr_read_inval(), pr_read_as(), pr_read_status(),
590 	pr_read_lstatus(), pr_read_psinfo(), pr_read_lpsinfo(),
591 	pr_read_map(), pr_read_rmap(), pr_read_xmap(),
592 	pr_read_cred(), pr_read_sigact(), pr_read_auxv(),
593 #if defined(__x86)
594 	pr_read_ldt(),
595 #endif
596 	pr_read_usage(), pr_read_lusage(), pr_read_pagedata(),
597 	pr_read_watch(), pr_read_lwpstatus(), pr_read_lwpsinfo(),
598 	pr_read_lwpusage(), pr_read_lwpname(),
599 	pr_read_xregs(), pr_read_priv(),
600 	pr_read_spymaster(), pr_read_secflags(),
601 #if defined(__sparc)
602 	pr_read_gwindows(), pr_read_asrs(),
603 #endif
604 	pr_read_piddir(), pr_read_pidfile(), pr_read_opagedata(),
605 	pr_read_fdinfo();
606 
607 static int (*pr_read_function[PR_NFILES])() = {
608 	pr_read_inval,		/* /proc				*/
609 	pr_read_inval,		/* /proc/self				*/
610 	pr_read_piddir,		/* /proc/<pid> (old /proc read())	*/
611 	pr_read_as,		/* /proc/<pid>/as			*/
612 	pr_read_inval,		/* /proc/<pid>/ctl			*/
613 	pr_read_status,		/* /proc/<pid>/status			*/
614 	pr_read_lstatus,	/* /proc/<pid>/lstatus			*/
615 	pr_read_psinfo,		/* /proc/<pid>/psinfo			*/
616 	pr_read_lpsinfo,	/* /proc/<pid>/lpsinfo			*/
617 	pr_read_map,		/* /proc/<pid>/map			*/
618 	pr_read_rmap,		/* /proc/<pid>/rmap			*/
619 	pr_read_xmap,		/* /proc/<pid>/xmap			*/
620 	pr_read_cred,		/* /proc/<pid>/cred			*/
621 	pr_read_sigact,		/* /proc/<pid>/sigact			*/
622 	pr_read_auxv,		/* /proc/<pid>/auxv			*/
623 #if defined(__x86)
624 	pr_read_ldt,		/* /proc/<pid>/ldt			*/
625 #endif
626 	pr_read_usage,		/* /proc/<pid>/usage			*/
627 	pr_read_lusage,		/* /proc/<pid>/lusage			*/
628 	pr_read_pagedata,	/* /proc/<pid>/pagedata			*/
629 	pr_read_watch,		/* /proc/<pid>/watch			*/
630 	pr_read_inval,		/* /proc/<pid>/cwd			*/
631 	pr_read_inval,		/* /proc/<pid>/root			*/
632 	pr_read_inval,		/* /proc/<pid>/fd			*/
633 	pr_read_inval,		/* /proc/<pid>/fd/nn			*/
634 	pr_read_inval,		/* /proc/<pid>/fdinfo			*/
635 	pr_read_fdinfo,		/* /proc/<pid>/fdinfo/nn		*/
636 	pr_read_inval,		/* /proc/<pid>/object			*/
637 	pr_read_inval,		/* /proc/<pid>/object/xxx		*/
638 	pr_read_inval,		/* /proc/<pid>/lwp			*/
639 	pr_read_inval,		/* /proc/<pid>/lwp/<lwpid>		*/
640 	pr_read_inval,		/* /proc/<pid>/lwp/<lwpid>/lwpctl	*/
641 	pr_read_lwpname,	/* /proc/<pid>/lwp/<lwpid>/lwpname	*/
642 	pr_read_lwpstatus,	/* /proc/<pid>/lwp/<lwpid>/lwpstatus	*/
643 	pr_read_lwpsinfo,	/* /proc/<pid>/lwp/<lwpid>/lwpsinfo	*/
644 	pr_read_lwpusage,	/* /proc/<pid>/lwp/<lwpid>/lwpusage	*/
645 	pr_read_xregs,		/* /proc/<pid>/lwp/<lwpid>/xregs	*/
646 	pr_read_inval,		/* /proc/<pid>/lwp/<lwpid>/templates	*/
647 	pr_read_inval,		/* /proc/<pid>/lwp/<lwpid>/templates/<id> */
648 	pr_read_spymaster,	/* /proc/<pid>/lwp/<lwpid>/spymaster	*/
649 #if defined(__sparc)
650 	pr_read_gwindows,	/* /proc/<pid>/lwp/<lwpid>/gwindows	*/
651 	pr_read_asrs,		/* /proc/<pid>/lwp/<lwpid>/asrs		*/
652 #endif
653 	pr_read_priv,		/* /proc/<pid>/priv			*/
654 	pr_read_inval,		/* /proc/<pid>/path			*/
655 	pr_read_inval,		/* /proc/<pid>/path/xxx			*/
656 	pr_read_inval,		/* /proc/<pid>/contracts		*/
657 	pr_read_inval,		/* /proc/<pid>/contracts/<ctid>		*/
658 	pr_read_secflags,	/* /proc/<pid>/secflags			*/
659 	pr_read_pidfile,	/* old process file			*/
660 	pr_read_pidfile,	/* old lwp file				*/
661 	pr_read_opagedata,	/* old pagedata file			*/
662 };
663 
664 /* ARGSUSED */
665 static int
666 pr_read_inval(prnode_t *pnp, uio_t *uiop, cred_t *cr)
667 {
668 	/*
669 	 * No read() on any /proc directory, use getdents(2) instead.
670 	 * Cannot read a control file either.
671 	 * An underlying mapped object file cannot get here.
672 	 */
673 	return (EINVAL);
674 }
675 
676 static int
677 pr_uioread(void *base, long count, uio_t *uiop)
678 {
679 	int error = 0;
680 
681 	ASSERT(count >= 0);
682 	count -= uiop->uio_offset;
683 	if (count > 0 && uiop->uio_offset >= 0) {
684 		error = uiomove((char *)base + uiop->uio_offset,
685 		    count, UIO_READ, uiop);
686 	}
687 
688 	return (error);
689 }
690 
691 static int
692 pr_read_as(prnode_t *pnp, uio_t *uiop)
693 {
694 	int error;
695 
696 	ASSERT(pnp->pr_type == PR_AS);
697 
698 	if ((error = prlock(pnp, ZNO)) == 0) {
699 		proc_t *p = pnp->pr_common->prc_proc;
700 		struct as *as = p->p_as;
701 
702 		/*
703 		 * /proc I/O cannot be done to a system process.
704 		 * A 32-bit process cannot read a 64-bit process.
705 		 */
706 		if ((p->p_flag & SSYS) || as == &kas) {
707 			error = 0;
708 #ifdef _SYSCALL32_IMPL
709 		} else if (curproc->p_model == DATAMODEL_ILP32 &&
710 		    PROCESS_NOT_32BIT(p)) {
711 			error = EOVERFLOW;
712 #endif
713 		} else {
714 			/*
715 			 * We don't hold p_lock over an i/o operation because
716 			 * that could lead to deadlock with the clock thread.
717 			 */
718 			mutex_exit(&p->p_lock);
719 			error = prusrio(p, UIO_READ, uiop, 0);
720 			mutex_enter(&p->p_lock);
721 		}
722 		prunlock(pnp);
723 	}
724 
725 	return (error);
726 }
727 
728 static int
729 pr_read_status(prnode_t *pnp, uio_t *uiop, cred_t *cr)
730 {
731 	pstatus_t *sp;
732 	int error;
733 
734 	ASSERT(pnp->pr_type == PR_STATUS);
735 
736 	/*
737 	 * We kmem_alloc() the pstatus structure because
738 	 * it is so big it might blow the kernel stack.
739 	 */
740 	sp = kmem_alloc(sizeof (*sp), KM_SLEEP);
741 	if ((error = prlock(pnp, ZNO)) == 0) {
742 		prgetstatus(pnp->pr_common->prc_proc, sp, VTOZONE(PTOV(pnp)));
743 		prunlock(pnp);
744 		error = pr_uioread(sp, sizeof (*sp), uiop);
745 	}
746 	kmem_free(sp, sizeof (*sp));
747 	return (error);
748 }
749 
750 static int
751 pr_read_lstatus(prnode_t *pnp, uio_t *uiop, cred_t *cr)
752 {
753 	proc_t *p;
754 	kthread_t *t;
755 	lwpdir_t *ldp;
756 	size_t size;
757 	prheader_t *php;
758 	lwpstatus_t *sp;
759 	int error;
760 	int nlwp;
761 	int i;
762 
763 	ASSERT(pnp->pr_type == PR_LSTATUS);
764 
765 	if ((error = prlock(pnp, ZNO)) != 0)
766 		return (error);
767 	p = pnp->pr_common->prc_proc;
768 	nlwp = p->p_lwpcnt;
769 	size = sizeof (prheader_t) + nlwp * LSPAN(lwpstatus_t);
770 
771 	/* drop p->p_lock to do kmem_alloc(KM_SLEEP) */
772 	mutex_exit(&p->p_lock);
773 	php = kmem_zalloc(size, KM_SLEEP);
774 	mutex_enter(&p->p_lock);
775 	/* p->p_lwpcnt can't change while process is locked */
776 	ASSERT(nlwp == p->p_lwpcnt);
777 
778 	php->pr_nent = nlwp;
779 	php->pr_entsize = LSPAN(lwpstatus_t);
780 
781 	sp = (lwpstatus_t *)(php + 1);
782 	for (ldp = p->p_lwpdir, i = 0; i < p->p_lwpdir_sz; i++, ldp++) {
783 		if (ldp->ld_entry == NULL ||
784 		    (t = ldp->ld_entry->le_thread) == NULL)
785 			continue;
786 		prgetlwpstatus(t, sp, VTOZONE(PTOV(pnp)));
787 		sp = (lwpstatus_t *)((caddr_t)sp + LSPAN(lwpstatus_t));
788 	}
789 	prunlock(pnp);
790 
791 	error = pr_uioread(php, size, uiop);
792 	kmem_free(php, size);
793 	return (error);
794 }
795 
796 static int
797 pr_read_psinfo(prnode_t *pnp, uio_t *uiop, cred_t *cr)
798 {
799 	psinfo_t psinfo;
800 	proc_t *p;
801 	int error = 0;
802 
803 	ASSERT(pnp->pr_type == PR_PSINFO);
804 
805 	/*
806 	 * We don't want the full treatment of prlock(pnp) here.
807 	 * This file is world-readable and never goes invalid.
808 	 * It doesn't matter if we are in the middle of an exec().
809 	 */
810 	p = pr_p_lock(pnp);
811 	mutex_exit(&pr_pidlock);
812 	if (p == NULL)
813 		error = ENOENT;
814 	else {
815 		ASSERT(p == pnp->pr_common->prc_proc);
816 		prgetpsinfo(p, &psinfo);
817 		prunlock(pnp);
818 		error = pr_uioread(&psinfo, sizeof (psinfo), uiop);
819 	}
820 	return (error);
821 }
822 
823 static int
824 pr_read_fdinfo(prnode_t *pnp, uio_t *uiop, cred_t *cr)
825 {
826 	prfdinfo_t *fdinfo;
827 	list_t data;
828 	proc_t *p;
829 	uint_t fd;
830 	file_t *fp;
831 	short ufp_flag;
832 	int error = 0;
833 
834 	ASSERT(pnp->pr_type == PR_FDINFO);
835 
836 	/*
837 	 * This is a guess at the size of the structure that needs to
838 	 * be returned. It's a balance between not allocating too much more
839 	 * space than is required and not requiring too many subsequent
840 	 * reallocations. Allocate it before acquiring the process lock.
841 	 */
842 	pr_iol_initlist(&data, sizeof (prfdinfo_t) + MAXPATHLEN + 2, 1);
843 
844 	if ((error = prlock(pnp, ZNO)) != 0) {
845 		pr_iol_freelist(&data);
846 		return (error);
847 	}
848 
849 	p = pnp->pr_common->prc_proc;
850 
851 	if ((p->p_flag & SSYS) || p->p_as == &kas) {
852 		prunlock(pnp);
853 		pr_iol_freelist(&data);
854 		return (0);
855 	}
856 
857 	fd = pnp->pr_index;
858 
859 	/* Fetch and lock the file_t for this descriptor */
860 	fp = pr_getf(p, fd, &ufp_flag);
861 
862 	if (fp == NULL) {
863 		error = ENOENT;
864 		prunlock(pnp);
865 		goto out;
866 	}
867 
868 	/*
869 	 * For fdinfo, we don't want to include the placeholder pr_misc at the
870 	 * end of the struct. We'll terminate the data with an empty pr_misc
871 	 * header before returning.
872 	 */
873 
874 	fdinfo = pr_iol_newbuf(&data, offsetof(prfdinfo_t, pr_misc));
875 	fdinfo->pr_fd = fd;
876 	fdinfo->pr_fdflags = ufp_flag;
877 	fdinfo->pr_fileflags = fp->f_flag2 << 16 | fp->f_flag;
878 	if ((fdinfo->pr_fileflags & (FSEARCH | FEXEC)) == 0)
879 		fdinfo->pr_fileflags += FOPEN;
880 	fdinfo->pr_offset = fp->f_offset;
881 	/*
882 	 * Information from the vnode (rather than the file_t) is retrieved
883 	 * later, in prgetfdinfo() - for example sock_getfasync()
884 	 */
885 
886 	prunlock(pnp);
887 
888 	error = prgetfdinfo(p, fp->f_vnode, fdinfo, cr, fp->f_cred, &data);
889 
890 	pr_releasef(fp);
891 
892 out:
893 	if (error == 0)
894 		error = pr_iol_uiomove_and_free(&data, uiop, error);
895 	else
896 		pr_iol_freelist(&data);
897 
898 	return (error);
899 }
900 
901 static int
902 pr_read_lpsinfo(prnode_t *pnp, uio_t *uiop, cred_t *cr)
903 {
904 	proc_t *p;
905 	kthread_t *t;
906 	lwpdir_t *ldp;
907 	lwpent_t *lep;
908 	size_t size;
909 	prheader_t *php;
910 	lwpsinfo_t *sp;
911 	int error;
912 	int nlwp;
913 	int i;
914 
915 	ASSERT(pnp->pr_type == PR_LPSINFO);
916 
917 	/*
918 	 * We don't want the full treatment of prlock(pnp) here.
919 	 * This file is world-readable and never goes invalid.
920 	 * It doesn't matter if we are in the middle of an exec().
921 	 */
922 	p = pr_p_lock(pnp);
923 	mutex_exit(&pr_pidlock);
924 	if (p == NULL)
925 		return (ENOENT);
926 	ASSERT(p == pnp->pr_common->prc_proc);
927 	if ((nlwp = p->p_lwpcnt + p->p_zombcnt) == 0) {
928 		prunlock(pnp);
929 		return (ENOENT);
930 	}
931 	size = sizeof (prheader_t) + nlwp * LSPAN(lwpsinfo_t);
932 
933 	/* drop p->p_lock to do kmem_alloc(KM_SLEEP) */
934 	mutex_exit(&p->p_lock);
935 	php = kmem_zalloc(size, KM_SLEEP);
936 	mutex_enter(&p->p_lock);
937 	/* p->p_lwpcnt can't change while process is locked */
938 	ASSERT(nlwp == p->p_lwpcnt + p->p_zombcnt);
939 
940 	php->pr_nent = nlwp;
941 	php->pr_entsize = LSPAN(lwpsinfo_t);
942 
943 	sp = (lwpsinfo_t *)(php + 1);
944 	for (ldp = p->p_lwpdir, i = 0; i < p->p_lwpdir_sz; i++, ldp++) {
945 		if ((lep = ldp->ld_entry) == NULL)
946 			continue;
947 		if ((t = lep->le_thread) != NULL)
948 			prgetlwpsinfo(t, sp);
949 		else {
950 			bzero(sp, sizeof (*sp));
951 			sp->pr_lwpid = lep->le_lwpid;
952 			sp->pr_state = SZOMB;
953 			sp->pr_sname = 'Z';
954 			sp->pr_start.tv_sec = lep->le_start;
955 			sp->pr_bindpro = PBIND_NONE;
956 			sp->pr_bindpset = PS_NONE;
957 		}
958 		sp = (lwpsinfo_t *)((caddr_t)sp + LSPAN(lwpsinfo_t));
959 	}
960 	prunlock(pnp);
961 
962 	error = pr_uioread(php, size, uiop);
963 	kmem_free(php, size);
964 	return (error);
965 }
966 
967 static int
968 pr_read_map_common(prnode_t *pnp, uio_t *uiop, prnodetype_t type)
969 {
970 	proc_t *p;
971 	struct as *as;
972 	list_t iolhead;
973 	int error;
974 
975 readmap_common:
976 	if ((error = prlock(pnp, ZNO)) != 0)
977 		return (error);
978 
979 	p = pnp->pr_common->prc_proc;
980 	as = p->p_as;
981 
982 	if ((p->p_flag & SSYS) || as == &kas) {
983 		prunlock(pnp);
984 		return (0);
985 	}
986 
987 	if (!AS_LOCK_TRYENTER(as, RW_WRITER)) {
988 		prunlock(pnp);
989 		delay(1);
990 		goto readmap_common;
991 	}
992 	mutex_exit(&p->p_lock);
993 
994 	switch (type) {
995 	case PR_XMAP:
996 		error = prgetxmap(p, &iolhead);
997 		break;
998 	case PR_RMAP:
999 		error = prgetmap(p, 1, &iolhead);
1000 		break;
1001 	case PR_MAP:
1002 		error = prgetmap(p, 0, &iolhead);
1003 		break;
1004 	}
1005 
1006 	AS_LOCK_EXIT(as);
1007 	mutex_enter(&p->p_lock);
1008 	prunlock(pnp);
1009 
1010 	error = pr_iol_uiomove_and_free(&iolhead, uiop, error);
1011 
1012 	return (error);
1013 }
1014 
1015 static int
1016 pr_read_map(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1017 {
1018 	ASSERT(pnp->pr_type == PR_MAP);
1019 	return (pr_read_map_common(pnp, uiop, pnp->pr_type));
1020 }
1021 
1022 static int
1023 pr_read_rmap(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1024 {
1025 	ASSERT(pnp->pr_type == PR_RMAP);
1026 	return (pr_read_map_common(pnp, uiop, pnp->pr_type));
1027 }
1028 
1029 static int
1030 pr_read_xmap(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1031 {
1032 	ASSERT(pnp->pr_type == PR_XMAP);
1033 	return (pr_read_map_common(pnp, uiop, pnp->pr_type));
1034 }
1035 
1036 static int
1037 pr_read_cred(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1038 {
1039 	proc_t *p;
1040 	prcred_t *pcrp;
1041 	int error;
1042 	size_t count;
1043 
1044 	ASSERT(pnp->pr_type == PR_CRED);
1045 
1046 	/*
1047 	 * We kmem_alloc() the prcred_t structure because
1048 	 * the number of supplementary groups is variable.
1049 	 */
1050 	pcrp =
1051 	    kmem_alloc(sizeof (prcred_t) + sizeof (gid_t) * (ngroups_max - 1),
1052 	    KM_SLEEP);
1053 
1054 	if ((error = prlock(pnp, ZNO)) != 0)
1055 		goto out;
1056 	p = pnp->pr_common->prc_proc;
1057 	ASSERT(p != NULL);
1058 
1059 	prgetcred(p, pcrp);
1060 	prunlock(pnp);
1061 
1062 	count = sizeof (prcred_t);
1063 	if (pcrp->pr_ngroups > 1)
1064 		count += sizeof (gid_t) * (pcrp->pr_ngroups - 1);
1065 	error = pr_uioread(pcrp, count, uiop);
1066 out:
1067 	kmem_free(pcrp, sizeof (prcred_t) + sizeof (gid_t) * (ngroups_max - 1));
1068 	return (error);
1069 }
1070 
1071 static int
1072 pr_read_priv(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1073 {
1074 	proc_t *p;
1075 	size_t psize = prgetprivsize();
1076 	prpriv_t *ppriv = kmem_alloc(psize, KM_SLEEP);
1077 	int error;
1078 
1079 	ASSERT(pnp->pr_type == PR_PRIV);
1080 
1081 	if ((error = prlock(pnp, ZNO)) != 0)
1082 		goto out;
1083 	p = pnp->pr_common->prc_proc;
1084 	ASSERT(p != NULL);
1085 
1086 	prgetpriv(p, ppriv);
1087 	prunlock(pnp);
1088 
1089 	error = pr_uioread(ppriv, psize, uiop);
1090 out:
1091 	kmem_free(ppriv, psize);
1092 	return (error);
1093 }
1094 
1095 static int
1096 pr_read_sigact(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1097 {
1098 	int nsig = PROC_IS_BRANDED(curproc)? BROP(curproc)->b_nsig : NSIG;
1099 	proc_t *p;
1100 	struct sigaction *sap;
1101 	int sig;
1102 	int error;
1103 	user_t *up;
1104 
1105 	ASSERT(pnp->pr_type == PR_SIGACT);
1106 
1107 	/*
1108 	 * We kmem_alloc() the sigaction array because
1109 	 * it is so big it might blow the kernel stack.
1110 	 */
1111 	sap = kmem_alloc((nsig-1) * sizeof (struct sigaction), KM_SLEEP);
1112 
1113 	if ((error = prlock(pnp, ZNO)) != 0)
1114 		goto out;
1115 	p = pnp->pr_common->prc_proc;
1116 	ASSERT(p != NULL);
1117 
1118 	if (uiop->uio_offset >= (nsig-1)*sizeof (struct sigaction)) {
1119 		prunlock(pnp);
1120 		goto out;
1121 	}
1122 
1123 	up = PTOU(p);
1124 	for (sig = 1; sig < nsig; sig++)
1125 		prgetaction(p, up, sig, &sap[sig-1]);
1126 	prunlock(pnp);
1127 
1128 	error = pr_uioread(sap, (nsig - 1) * sizeof (struct sigaction), uiop);
1129 out:
1130 	kmem_free(sap, (nsig-1) * sizeof (struct sigaction));
1131 	return (error);
1132 }
1133 
1134 static int
1135 pr_read_auxv(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1136 {
1137 	auxv_t auxv[__KERN_NAUXV_IMPL];
1138 	proc_t *p;
1139 	user_t *up;
1140 	int error;
1141 
1142 	ASSERT(pnp->pr_type == PR_AUXV);
1143 
1144 	if ((error = prlock(pnp, ZNO)) != 0)
1145 		return (error);
1146 
1147 	if (uiop->uio_offset >= sizeof (auxv)) {
1148 		prunlock(pnp);
1149 		return (0);
1150 	}
1151 
1152 	p = pnp->pr_common->prc_proc;
1153 	up = PTOU(p);
1154 	bcopy(up->u_auxv, auxv, sizeof (auxv));
1155 	prunlock(pnp);
1156 
1157 	return (pr_uioread(auxv, sizeof (auxv), uiop));
1158 }
1159 
1160 #if defined(__x86)
1161 /*
1162  * XX64
1163  *	This is almost certainly broken for the amd64 kernel, because
1164  *	we have two kinds of LDT structures to export -- one for compatibility
1165  *	mode, and one for long mode, sigh.
1166  *
1167  *	For now let's just have a ldt of size 0 for 64-bit processes.
1168  */
1169 static int
1170 pr_read_ldt(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1171 {
1172 	proc_t *p;
1173 	struct ssd *ssd;
1174 	size_t size;
1175 	int error;
1176 
1177 	ASSERT(pnp->pr_type == PR_LDT);
1178 
1179 	if ((error = prlock(pnp, ZNO)) != 0)
1180 		return (error);
1181 	p = pnp->pr_common->prc_proc;
1182 
1183 	mutex_exit(&p->p_lock);
1184 	mutex_enter(&p->p_ldtlock);
1185 	size = prnldt(p) * sizeof (struct ssd);
1186 	if (uiop->uio_offset >= size) {
1187 		mutex_exit(&p->p_ldtlock);
1188 		mutex_enter(&p->p_lock);
1189 		prunlock(pnp);
1190 		return (0);
1191 	}
1192 
1193 	ssd = kmem_alloc(size, KM_SLEEP);
1194 	prgetldt(p, ssd);
1195 	mutex_exit(&p->p_ldtlock);
1196 	mutex_enter(&p->p_lock);
1197 	prunlock(pnp);
1198 
1199 	error = pr_uioread(ssd, size, uiop);
1200 	kmem_free(ssd, size);
1201 	return (error);
1202 }
1203 #endif	/* __x86 */
1204 
1205 static int
1206 pr_read_usage(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1207 {
1208 	prhusage_t *pup;
1209 	prusage_t *upup;
1210 	proc_t *p;
1211 	kthread_t *t;
1212 	int error;
1213 
1214 	ASSERT(pnp->pr_type == PR_USAGE);
1215 
1216 	/* allocate now, before locking the process */
1217 	pup = kmem_zalloc(sizeof (*pup), KM_SLEEP);
1218 	upup = kmem_alloc(sizeof (*upup), KM_SLEEP);
1219 
1220 	/*
1221 	 * We don't want the full treatment of prlock(pnp) here.
1222 	 * This file is world-readable and never goes invalid.
1223 	 * It doesn't matter if we are in the middle of an exec().
1224 	 */
1225 	p = pr_p_lock(pnp);
1226 	mutex_exit(&pr_pidlock);
1227 	if (p == NULL) {
1228 		error = ENOENT;
1229 		goto out;
1230 	}
1231 	ASSERT(p == pnp->pr_common->prc_proc);
1232 
1233 	if (uiop->uio_offset >= sizeof (prusage_t)) {
1234 		prunlock(pnp);
1235 		error = 0;
1236 		goto out;
1237 	}
1238 
1239 	pup->pr_tstamp = gethrtime();
1240 
1241 	pup->pr_count  = p->p_defunct;
1242 	pup->pr_create = p->p_mstart;
1243 	pup->pr_term   = p->p_mterm;
1244 
1245 	pup->pr_rtime    = p->p_mlreal;
1246 	pup->pr_utime    = p->p_acct[LMS_USER];
1247 	pup->pr_stime    = p->p_acct[LMS_SYSTEM];
1248 	pup->pr_ttime    = p->p_acct[LMS_TRAP];
1249 	pup->pr_tftime   = p->p_acct[LMS_TFAULT];
1250 	pup->pr_dftime   = p->p_acct[LMS_DFAULT];
1251 	pup->pr_kftime   = p->p_acct[LMS_KFAULT];
1252 	pup->pr_ltime    = p->p_acct[LMS_USER_LOCK];
1253 	pup->pr_slptime  = p->p_acct[LMS_SLEEP];
1254 	pup->pr_wtime    = p->p_acct[LMS_WAIT_CPU];
1255 	pup->pr_stoptime = p->p_acct[LMS_STOPPED];
1256 
1257 	pup->pr_minf  = p->p_ru.minflt;
1258 	pup->pr_majf  = p->p_ru.majflt;
1259 	pup->pr_nswap = p->p_ru.nswap;
1260 	pup->pr_inblk = p->p_ru.inblock;
1261 	pup->pr_oublk = p->p_ru.oublock;
1262 	pup->pr_msnd  = p->p_ru.msgsnd;
1263 	pup->pr_mrcv  = p->p_ru.msgrcv;
1264 	pup->pr_sigs  = p->p_ru.nsignals;
1265 	pup->pr_vctx  = p->p_ru.nvcsw;
1266 	pup->pr_ictx  = p->p_ru.nivcsw;
1267 	pup->pr_sysc  = p->p_ru.sysc;
1268 	pup->pr_ioch  = p->p_ru.ioch;
1269 
1270 	/*
1271 	 * Add the usage information for each active lwp.
1272 	 */
1273 	if ((t = p->p_tlist) != NULL &&
1274 	    !(pnp->pr_pcommon->prc_flags & PRC_DESTROY)) {
1275 		do {
1276 			if (t->t_proc_flag & TP_LWPEXIT)
1277 				continue;
1278 			pup->pr_count++;
1279 			praddusage(t, pup);
1280 		} while ((t = t->t_forw) != p->p_tlist);
1281 	}
1282 
1283 	prunlock(pnp);
1284 
1285 	prcvtusage(pup, upup);
1286 
1287 	error = pr_uioread(upup, sizeof (prusage_t), uiop);
1288 out:
1289 	kmem_free(pup, sizeof (*pup));
1290 	kmem_free(upup, sizeof (*upup));
1291 	return (error);
1292 }
1293 
1294 static int
1295 pr_read_lusage(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1296 {
1297 	int nlwp;
1298 	prhusage_t *pup;
1299 	prheader_t *php;
1300 	prusage_t *upup;
1301 	size_t size;
1302 	hrtime_t curtime;
1303 	proc_t *p;
1304 	kthread_t *t;
1305 	lwpdir_t *ldp;
1306 	int error;
1307 	int i;
1308 
1309 	ASSERT(pnp->pr_type == PR_LUSAGE);
1310 
1311 	/*
1312 	 * We don't want the full treatment of prlock(pnp) here.
1313 	 * This file is world-readable and never goes invalid.
1314 	 * It doesn't matter if we are in the middle of an exec().
1315 	 */
1316 	p = pr_p_lock(pnp);
1317 	mutex_exit(&pr_pidlock);
1318 	if (p == NULL)
1319 		return (ENOENT);
1320 	ASSERT(p == pnp->pr_common->prc_proc);
1321 	if ((nlwp = p->p_lwpcnt) == 0) {
1322 		prunlock(pnp);
1323 		return (ENOENT);
1324 	}
1325 
1326 	size = sizeof (prheader_t) + (nlwp + 1) * LSPAN(prusage_t);
1327 	if (uiop->uio_offset >= size) {
1328 		prunlock(pnp);
1329 		return (0);
1330 	}
1331 
1332 	/* drop p->p_lock to do kmem_alloc(KM_SLEEP) */
1333 	mutex_exit(&p->p_lock);
1334 	pup = kmem_zalloc(size + sizeof (prhusage_t), KM_SLEEP);
1335 	mutex_enter(&p->p_lock);
1336 	/* p->p_lwpcnt can't change while process is locked */
1337 	ASSERT(nlwp == p->p_lwpcnt);
1338 
1339 	php = (prheader_t *)(pup + 1);
1340 	upup = (prusage_t *)(php + 1);
1341 
1342 	php->pr_nent = nlwp + 1;
1343 	php->pr_entsize = LSPAN(prusage_t);
1344 
1345 	curtime = gethrtime();
1346 
1347 	/*
1348 	 * First the summation over defunct lwps.
1349 	 */
1350 	pup->pr_count  = p->p_defunct;
1351 	pup->pr_tstamp = curtime;
1352 	pup->pr_create = p->p_mstart;
1353 	pup->pr_term   = p->p_mterm;
1354 
1355 	pup->pr_rtime    = p->p_mlreal;
1356 	pup->pr_utime    = p->p_acct[LMS_USER];
1357 	pup->pr_stime    = p->p_acct[LMS_SYSTEM];
1358 	pup->pr_ttime    = p->p_acct[LMS_TRAP];
1359 	pup->pr_tftime   = p->p_acct[LMS_TFAULT];
1360 	pup->pr_dftime   = p->p_acct[LMS_DFAULT];
1361 	pup->pr_kftime   = p->p_acct[LMS_KFAULT];
1362 	pup->pr_ltime    = p->p_acct[LMS_USER_LOCK];
1363 	pup->pr_slptime  = p->p_acct[LMS_SLEEP];
1364 	pup->pr_wtime    = p->p_acct[LMS_WAIT_CPU];
1365 	pup->pr_stoptime = p->p_acct[LMS_STOPPED];
1366 
1367 	pup->pr_minf  = p->p_ru.minflt;
1368 	pup->pr_majf  = p->p_ru.majflt;
1369 	pup->pr_nswap = p->p_ru.nswap;
1370 	pup->pr_inblk = p->p_ru.inblock;
1371 	pup->pr_oublk = p->p_ru.oublock;
1372 	pup->pr_msnd  = p->p_ru.msgsnd;
1373 	pup->pr_mrcv  = p->p_ru.msgrcv;
1374 	pup->pr_sigs  = p->p_ru.nsignals;
1375 	pup->pr_vctx  = p->p_ru.nvcsw;
1376 	pup->pr_ictx  = p->p_ru.nivcsw;
1377 	pup->pr_sysc  = p->p_ru.sysc;
1378 	pup->pr_ioch  = p->p_ru.ioch;
1379 
1380 	prcvtusage(pup, upup);
1381 
1382 	/*
1383 	 * Fill one prusage struct for each active lwp.
1384 	 */
1385 	for (ldp = p->p_lwpdir, i = 0; i < p->p_lwpdir_sz; i++, ldp++) {
1386 		if (ldp->ld_entry == NULL ||
1387 		    (t = ldp->ld_entry->le_thread) == NULL)
1388 			continue;
1389 		ASSERT(!(t->t_proc_flag & TP_LWPEXIT));
1390 		ASSERT(nlwp > 0);
1391 		--nlwp;
1392 		upup = (prusage_t *)((caddr_t)upup + LSPAN(prusage_t));
1393 		prgetusage(t, pup);
1394 		prcvtusage(pup, upup);
1395 	}
1396 	ASSERT(nlwp == 0);
1397 
1398 	prunlock(pnp);
1399 
1400 	error = pr_uioread(php, size, uiop);
1401 	kmem_free(pup, size + sizeof (prhusage_t));
1402 	return (error);
1403 }
1404 
1405 static int
1406 pr_read_pagedata(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1407 {
1408 	proc_t *p;
1409 	int error;
1410 
1411 	ASSERT(pnp->pr_type == PR_PAGEDATA);
1412 
1413 	if ((error = prlock(pnp, ZNO)) != 0)
1414 		return (error);
1415 
1416 	p = pnp->pr_common->prc_proc;
1417 	if ((p->p_flag & SSYS) || p->p_as == &kas) {
1418 		prunlock(pnp);
1419 		return (0);
1420 	}
1421 
1422 	mutex_exit(&p->p_lock);
1423 	error = prpdread(p, pnp->pr_hatid, uiop);
1424 	mutex_enter(&p->p_lock);
1425 
1426 	prunlock(pnp);
1427 	return (error);
1428 }
1429 
1430 static int
1431 pr_read_opagedata(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1432 {
1433 	proc_t *p;
1434 	struct as *as;
1435 	int error;
1436 
1437 	ASSERT(pnp->pr_type == PR_OPAGEDATA);
1438 
1439 	if ((error = prlock(pnp, ZNO)) != 0)
1440 		return (error);
1441 
1442 	p = pnp->pr_common->prc_proc;
1443 	as = p->p_as;
1444 	if ((p->p_flag & SSYS) || as == &kas) {
1445 		prunlock(pnp);
1446 		return (0);
1447 	}
1448 
1449 	mutex_exit(&p->p_lock);
1450 	error = oprpdread(as, pnp->pr_hatid, uiop);
1451 	mutex_enter(&p->p_lock);
1452 
1453 	prunlock(pnp);
1454 	return (error);
1455 }
1456 
1457 static int
1458 pr_read_watch(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1459 {
1460 	proc_t *p;
1461 	int error;
1462 	prwatch_t *Bpwp;
1463 	size_t size;
1464 	prwatch_t *pwp;
1465 	int nwarea;
1466 	struct watched_area *pwarea;
1467 
1468 	ASSERT(pnp->pr_type == PR_WATCH);
1469 
1470 	if ((error = prlock(pnp, ZNO)) != 0)
1471 		return (error);
1472 
1473 	p = pnp->pr_common->prc_proc;
1474 	nwarea = avl_numnodes(&p->p_warea);
1475 	size = nwarea * sizeof (prwatch_t);
1476 	if (uiop->uio_offset >= size) {
1477 		prunlock(pnp);
1478 		return (0);
1479 	}
1480 
1481 	/* drop p->p_lock to do kmem_alloc(KM_SLEEP) */
1482 	mutex_exit(&p->p_lock);
1483 	Bpwp = pwp = kmem_zalloc(size, KM_SLEEP);
1484 	mutex_enter(&p->p_lock);
1485 	/* p->p_nwarea can't change while process is locked */
1486 	ASSERT(nwarea == avl_numnodes(&p->p_warea));
1487 
1488 	/* gather the watched areas */
1489 	for (pwarea = avl_first(&p->p_warea); pwarea != NULL;
1490 	    pwarea = AVL_NEXT(&p->p_warea, pwarea), pwp++) {
1491 		pwp->pr_vaddr = (uintptr_t)pwarea->wa_vaddr;
1492 		pwp->pr_size = pwarea->wa_eaddr - pwarea->wa_vaddr;
1493 		pwp->pr_wflags = (int)pwarea->wa_flags;
1494 	}
1495 
1496 	prunlock(pnp);
1497 
1498 	error = pr_uioread(Bpwp, size, uiop);
1499 	kmem_free(Bpwp, size);
1500 	return (error);
1501 }
1502 
1503 static int
1504 pr_read_lwpstatus(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1505 {
1506 	lwpstatus_t *sp;
1507 	int error;
1508 
1509 	ASSERT(pnp->pr_type == PR_LWPSTATUS);
1510 
1511 	/*
1512 	 * We kmem_alloc() the lwpstatus structure because
1513 	 * it is so big it might blow the kernel stack.
1514 	 */
1515 	sp = kmem_alloc(sizeof (*sp), KM_SLEEP);
1516 
1517 	if ((error = prlock(pnp, ZNO)) != 0)
1518 		goto out;
1519 
1520 	if (uiop->uio_offset >= sizeof (*sp)) {
1521 		prunlock(pnp);
1522 		goto out;
1523 	}
1524 
1525 	prgetlwpstatus(pnp->pr_common->prc_thread, sp, VTOZONE(PTOV(pnp)));
1526 	prunlock(pnp);
1527 
1528 	error = pr_uioread(sp, sizeof (*sp), uiop);
1529 out:
1530 	kmem_free(sp, sizeof (*sp));
1531 	return (error);
1532 }
1533 
1534 static int
1535 pr_read_lwpsinfo(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1536 {
1537 	lwpsinfo_t lwpsinfo;
1538 	proc_t *p;
1539 	kthread_t *t;
1540 	lwpent_t *lep;
1541 
1542 	ASSERT(pnp->pr_type == PR_LWPSINFO);
1543 
1544 	/*
1545 	 * We don't want the full treatment of prlock(pnp) here.
1546 	 * This file is world-readable and never goes invalid.
1547 	 * It doesn't matter if we are in the middle of an exec().
1548 	 */
1549 	p = pr_p_lock(pnp);
1550 	mutex_exit(&pr_pidlock);
1551 	if (p == NULL)
1552 		return (ENOENT);
1553 	ASSERT(p == pnp->pr_common->prc_proc);
1554 	if (pnp->pr_common->prc_tslot == -1) {
1555 		prunlock(pnp);
1556 		return (ENOENT);
1557 	}
1558 
1559 	if (uiop->uio_offset >= sizeof (lwpsinfo)) {
1560 		prunlock(pnp);
1561 		return (0);
1562 	}
1563 
1564 	if ((t = pnp->pr_common->prc_thread) != NULL)
1565 		prgetlwpsinfo(t, &lwpsinfo);
1566 	else {
1567 		lep = p->p_lwpdir[pnp->pr_common->prc_tslot].ld_entry;
1568 		bzero(&lwpsinfo, sizeof (lwpsinfo));
1569 		lwpsinfo.pr_lwpid = lep->le_lwpid;
1570 		lwpsinfo.pr_state = SZOMB;
1571 		lwpsinfo.pr_sname = 'Z';
1572 		lwpsinfo.pr_start.tv_sec = lep->le_start;
1573 		lwpsinfo.pr_bindpro = PBIND_NONE;
1574 		lwpsinfo.pr_bindpset = PS_NONE;
1575 	}
1576 	prunlock(pnp);
1577 
1578 	return (pr_uioread(&lwpsinfo, sizeof (lwpsinfo), uiop));
1579 }
1580 
1581 static int
1582 pr_read_lwpusage(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1583 {
1584 	prhusage_t *pup;
1585 	prusage_t *upup;
1586 	proc_t *p;
1587 	int error;
1588 
1589 	ASSERT(pnp->pr_type == PR_LWPUSAGE);
1590 
1591 	/* allocate now, before locking the process */
1592 	pup = kmem_zalloc(sizeof (*pup), KM_SLEEP);
1593 	upup = kmem_alloc(sizeof (*upup), KM_SLEEP);
1594 
1595 	/*
1596 	 * We don't want the full treatment of prlock(pnp) here.
1597 	 * This file is world-readable and never goes invalid.
1598 	 * It doesn't matter if we are in the middle of an exec().
1599 	 */
1600 	p = pr_p_lock(pnp);
1601 	mutex_exit(&pr_pidlock);
1602 	if (p == NULL) {
1603 		error = ENOENT;
1604 		goto out;
1605 	}
1606 	ASSERT(p == pnp->pr_common->prc_proc);
1607 	if (pnp->pr_common->prc_thread == NULL) {
1608 		prunlock(pnp);
1609 		error = ENOENT;
1610 		goto out;
1611 	}
1612 	if (uiop->uio_offset >= sizeof (prusage_t)) {
1613 		prunlock(pnp);
1614 		error = 0;
1615 		goto out;
1616 	}
1617 
1618 	pup->pr_tstamp = gethrtime();
1619 	prgetusage(pnp->pr_common->prc_thread, pup);
1620 
1621 	prunlock(pnp);
1622 
1623 	prcvtusage(pup, upup);
1624 
1625 	error = pr_uioread(upup, sizeof (prusage_t), uiop);
1626 out:
1627 	kmem_free(pup, sizeof (*pup));
1628 	kmem_free(upup, sizeof (*upup));
1629 	return (error);
1630 }
1631 
1632 static int
1633 pr_read_lwpname(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1634 {
1635 	char lwpname[THREAD_NAME_MAX];
1636 	kthread_t *t;
1637 	int error;
1638 
1639 	ASSERT(pnp->pr_type == PR_LWPNAME);
1640 
1641 	if (uiop->uio_offset >= THREAD_NAME_MAX)
1642 		return (0);
1643 
1644 	if ((error = prlock(pnp, ZNO)) != 0)
1645 		return (error);
1646 
1647 	bzero(lwpname, sizeof (lwpname));
1648 
1649 	t = pnp->pr_common->prc_thread;
1650 
1651 	if (t->t_name != NULL)
1652 		(void) strlcpy(lwpname, t->t_name, sizeof (lwpname));
1653 
1654 	prunlock(pnp);
1655 
1656 	return (pr_uioread(lwpname, sizeof (lwpname), uiop));
1657 }
1658 
1659 /* ARGSUSED */
1660 static int
1661 pr_read_xregs(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1662 {
1663 #if defined(__sparc)
1664 	proc_t *p;
1665 	kthread_t *t;
1666 	int error;
1667 	char *xreg;
1668 	size_t size;
1669 
1670 	ASSERT(pnp->pr_type == PR_XREGS);
1671 
1672 	xreg = kmem_zalloc(sizeof (prxregset_t), KM_SLEEP);
1673 
1674 	if ((error = prlock(pnp, ZNO)) != 0)
1675 		goto out;
1676 
1677 	p = pnp->pr_common->prc_proc;
1678 	t = pnp->pr_common->prc_thread;
1679 
1680 	size = prhasx(p)? prgetprxregsize(p) : 0;
1681 	if (uiop->uio_offset >= size) {
1682 		prunlock(pnp);
1683 		goto out;
1684 	}
1685 
1686 	/* drop p->p_lock while (possibly) touching the stack */
1687 	mutex_exit(&p->p_lock);
1688 	prgetprxregs(ttolwp(t), xreg);
1689 	mutex_enter(&p->p_lock);
1690 	prunlock(pnp);
1691 
1692 	error = pr_uioread(xreg, size, uiop);
1693 out:
1694 	kmem_free(xreg, sizeof (prxregset_t));
1695 	return (error);
1696 #else
1697 	return (0);
1698 #endif
1699 }
1700 
1701 static int
1702 pr_read_spymaster(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1703 {
1704 	psinfo_t psinfo;
1705 	int error;
1706 	klwp_t *lwp;
1707 
1708 	ASSERT(pnp->pr_type == PR_SPYMASTER);
1709 
1710 	if ((error = prlock(pnp, ZNO)) != 0)
1711 		return (error);
1712 
1713 	if (pnp->pr_common->prc_thread == NULL) {
1714 		prunlock(pnp);
1715 		return (0);
1716 	}
1717 
1718 	lwp = pnp->pr_common->prc_thread->t_lwp;
1719 
1720 	if (lwp->lwp_spymaster == NULL) {
1721 		prunlock(pnp);
1722 		return (0);
1723 	}
1724 
1725 	bcopy(lwp->lwp_spymaster, &psinfo, sizeof (psinfo_t));
1726 	prunlock(pnp);
1727 
1728 	return (pr_uioread(&psinfo, sizeof (psinfo), uiop));
1729 }
1730 
1731 static int
1732 pr_read_secflags(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1733 {
1734 	prsecflags_t ret;
1735 	int error;
1736 	proc_t *p;
1737 
1738 	ASSERT(pnp->pr_type == PR_SECFLAGS);
1739 
1740 	if ((error = prlock(pnp, ZNO)) != 0)
1741 		return (error);
1742 
1743 	p = pnp->pr_common->prc_proc;
1744 	prgetsecflags(p, &ret);
1745 	prunlock(pnp);
1746 
1747 	return (pr_uioread(&ret, sizeof (ret), uiop));
1748 }
1749 
1750 #if defined(__sparc)
1751 
1752 static int
1753 pr_read_gwindows(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1754 {
1755 	proc_t *p;
1756 	kthread_t *t;
1757 	gwindows_t *gwp;
1758 	int error;
1759 	size_t size;
1760 
1761 	ASSERT(pnp->pr_type == PR_GWINDOWS);
1762 
1763 	gwp = kmem_zalloc(sizeof (gwindows_t), KM_SLEEP);
1764 
1765 	if ((error = prlock(pnp, ZNO)) != 0)
1766 		goto out;
1767 
1768 	p = pnp->pr_common->prc_proc;
1769 	t = pnp->pr_common->prc_thread;
1770 
1771 	/*
1772 	 * Drop p->p_lock while touching the stack.
1773 	 * The P_PR_LOCK flag prevents the lwp from
1774 	 * disappearing while we do this.
1775 	 */
1776 	mutex_exit(&p->p_lock);
1777 	if ((size = prnwindows(ttolwp(t))) != 0)
1778 		size = sizeof (gwindows_t) -
1779 		    (SPARC_MAXREGWINDOW - size) * sizeof (struct rwindow);
1780 	if (uiop->uio_offset >= size) {
1781 		mutex_enter(&p->p_lock);
1782 		prunlock(pnp);
1783 		goto out;
1784 	}
1785 	prgetwindows(ttolwp(t), gwp);
1786 	mutex_enter(&p->p_lock);
1787 	prunlock(pnp);
1788 
1789 	error = pr_uioread(gwp, size, uiop);
1790 out:
1791 	kmem_free(gwp, sizeof (gwindows_t));
1792 	return (error);
1793 }
1794 
1795 /* ARGSUSED */
1796 static int
1797 pr_read_asrs(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1798 {
1799 	int error;
1800 
1801 	ASSERT(pnp->pr_type == PR_ASRS);
1802 
1803 	/* the asrs file exists only for sparc v9 _LP64 processes */
1804 	if ((error = prlock(pnp, ZNO)) == 0) {
1805 		proc_t *p = pnp->pr_common->prc_proc;
1806 		kthread_t *t = pnp->pr_common->prc_thread;
1807 		asrset_t asrset;
1808 
1809 		if (p->p_model != DATAMODEL_LP64 ||
1810 		    uiop->uio_offset >= sizeof (asrset_t)) {
1811 			prunlock(pnp);
1812 			return (0);
1813 		}
1814 
1815 		/*
1816 		 * Drop p->p_lock while touching the stack.
1817 		 * The P_PR_LOCK flag prevents the lwp from
1818 		 * disappearing while we do this.
1819 		 */
1820 		mutex_exit(&p->p_lock);
1821 		prgetasregs(ttolwp(t), asrset);
1822 		mutex_enter(&p->p_lock);
1823 		prunlock(pnp);
1824 
1825 		error = pr_uioread(&asrset[0], sizeof (asrset_t), uiop);
1826 	}
1827 
1828 	return (error);
1829 }
1830 
1831 #endif	/* __sparc */
1832 
1833 static int
1834 pr_read_piddir(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1835 {
1836 	ASSERT(pnp->pr_type == PR_PIDDIR);
1837 	ASSERT(pnp->pr_pidfile != NULL);
1838 
1839 	/* use the underlying PR_PIDFILE to read the process */
1840 	pnp = VTOP(pnp->pr_pidfile);
1841 	ASSERT(pnp->pr_type == PR_PIDFILE);
1842 
1843 	return (pr_read_pidfile(pnp, uiop));
1844 }
1845 
1846 static int
1847 pr_read_pidfile(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1848 {
1849 	int error;
1850 
1851 	ASSERT(pnp->pr_type == PR_PIDFILE || pnp->pr_type == PR_LWPIDFILE);
1852 
1853 	if ((error = prlock(pnp, ZNO)) == 0) {
1854 		proc_t *p = pnp->pr_common->prc_proc;
1855 		struct as *as = p->p_as;
1856 
1857 		if ((p->p_flag & SSYS) || as == &kas) {
1858 			/*
1859 			 * /proc I/O cannot be done to a system process.
1860 			 */
1861 			error = EIO;	/* old /proc semantics */
1862 		} else {
1863 			/*
1864 			 * We drop p_lock because we don't want to hold
1865 			 * it over an I/O operation because that could
1866 			 * lead to deadlock with the clock thread.
1867 			 * The process will not disappear and its address
1868 			 * space will not change because it is marked P_PR_LOCK.
1869 			 */
1870 			mutex_exit(&p->p_lock);
1871 			error = prusrio(p, UIO_READ, uiop, 1);
1872 			mutex_enter(&p->p_lock);
1873 		}
1874 		prunlock(pnp);
1875 	}
1876 
1877 	return (error);
1878 }
1879 
1880 #ifdef _SYSCALL32_IMPL
1881 
1882 /*
1883  * Array of ILP32 read functions, indexed by /proc file type.
1884  */
1885 static int pr_read_status_32(),
1886 	pr_read_lstatus_32(), pr_read_psinfo_32(), pr_read_lpsinfo_32(),
1887 	pr_read_map_32(), pr_read_rmap_32(), pr_read_xmap_32(),
1888 	pr_read_sigact_32(), pr_read_auxv_32(),
1889 	pr_read_usage_32(), pr_read_lusage_32(), pr_read_pagedata_32(),
1890 	pr_read_watch_32(), pr_read_lwpstatus_32(), pr_read_lwpsinfo_32(),
1891 	pr_read_lwpusage_32(), pr_read_spymaster_32(),
1892 #if defined(__sparc)
1893 	pr_read_gwindows_32(),
1894 #endif
1895 	pr_read_opagedata_32();
1896 
1897 static int (*pr_read_function_32[PR_NFILES])() = {
1898 	pr_read_inval,		/* /proc				*/
1899 	pr_read_inval,		/* /proc/self				*/
1900 	pr_read_piddir,		/* /proc/<pid> (old /proc read())	*/
1901 	pr_read_as,		/* /proc/<pid>/as			*/
1902 	pr_read_inval,		/* /proc/<pid>/ctl			*/
1903 	pr_read_status_32,	/* /proc/<pid>/status			*/
1904 	pr_read_lstatus_32,	/* /proc/<pid>/lstatus			*/
1905 	pr_read_psinfo_32,	/* /proc/<pid>/psinfo			*/
1906 	pr_read_lpsinfo_32,	/* /proc/<pid>/lpsinfo			*/
1907 	pr_read_map_32,		/* /proc/<pid>/map			*/
1908 	pr_read_rmap_32,	/* /proc/<pid>/rmap			*/
1909 	pr_read_xmap_32,	/* /proc/<pid>/xmap			*/
1910 	pr_read_cred,		/* /proc/<pid>/cred			*/
1911 	pr_read_sigact_32,	/* /proc/<pid>/sigact			*/
1912 	pr_read_auxv_32,	/* /proc/<pid>/auxv			*/
1913 #if defined(__x86)
1914 	pr_read_ldt,		/* /proc/<pid>/ldt			*/
1915 #endif
1916 	pr_read_usage_32,	/* /proc/<pid>/usage			*/
1917 	pr_read_lusage_32,	/* /proc/<pid>/lusage			*/
1918 	pr_read_pagedata_32,	/* /proc/<pid>/pagedata			*/
1919 	pr_read_watch_32,	/* /proc/<pid>/watch			*/
1920 	pr_read_inval,		/* /proc/<pid>/cwd			*/
1921 	pr_read_inval,		/* /proc/<pid>/root			*/
1922 	pr_read_inval,		/* /proc/<pid>/fd			*/
1923 	pr_read_inval,		/* /proc/<pid>/fd/nn			*/
1924 	pr_read_inval,		/* /proc/<pid>/fdinfo			*/
1925 	pr_read_fdinfo,		/* /proc/<pid>/fdinfo/nn		*/
1926 	pr_read_inval,		/* /proc/<pid>/object			*/
1927 	pr_read_inval,		/* /proc/<pid>/object/xxx		*/
1928 	pr_read_inval,		/* /proc/<pid>/lwp			*/
1929 	pr_read_inval,		/* /proc/<pid>/lwp/<lwpid>		*/
1930 	pr_read_inval,		/* /proc/<pid>/lwp/<lwpid>/lwpctl	*/
1931 	pr_read_lwpname,	/* /proc/<pid>/lwp/<lwpid>/lwpname	*/
1932 	pr_read_lwpstatus_32,	/* /proc/<pid>/lwp/<lwpid>/lwpstatus	*/
1933 	pr_read_lwpsinfo_32,	/* /proc/<pid>/lwp/<lwpid>/lwpsinfo	*/
1934 	pr_read_lwpusage_32,	/* /proc/<pid>/lwp/<lwpid>/lwpusage	*/
1935 	pr_read_xregs,		/* /proc/<pid>/lwp/<lwpid>/xregs	*/
1936 	pr_read_inval,		/* /proc/<pid>/lwp/<lwpid>/templates	*/
1937 	pr_read_inval,		/* /proc/<pid>/lwp/<lwpid>/templates/<id> */
1938 	pr_read_spymaster_32,	/* /proc/<pid>/lwp/<lwpid>/spymaster	*/
1939 #if defined(__sparc)
1940 	pr_read_gwindows_32,	/* /proc/<pid>/lwp/<lwpid>/gwindows	*/
1941 	pr_read_asrs,		/* /proc/<pid>/lwp/<lwpid>/asrs		*/
1942 #endif
1943 	pr_read_priv,		/* /proc/<pid>/priv			*/
1944 	pr_read_inval,		/* /proc/<pid>/path			*/
1945 	pr_read_inval,		/* /proc/<pid>/path/xxx			*/
1946 	pr_read_inval,		/* /proc/<pid>/contracts		*/
1947 	pr_read_inval,		/* /proc/<pid>/contracts/<ctid>		*/
1948 	pr_read_secflags,	/* /proc/<pid>/secflags			*/
1949 	pr_read_pidfile,	/* old process file			*/
1950 	pr_read_pidfile,	/* old lwp file				*/
1951 	pr_read_opagedata_32,	/* old pagedata file			*/
1952 };
1953 
1954 static int
1955 pr_read_status_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1956 {
1957 	pstatus32_t *sp;
1958 	proc_t *p;
1959 	int error;
1960 
1961 	ASSERT(pnp->pr_type == PR_STATUS);
1962 
1963 	/*
1964 	 * We kmem_alloc() the pstatus structure because
1965 	 * it is so big it might blow the kernel stack.
1966 	 */
1967 	sp = kmem_alloc(sizeof (*sp), KM_SLEEP);
1968 	if ((error = prlock(pnp, ZNO)) == 0) {
1969 		/*
1970 		 * A 32-bit process cannot get the status of a 64-bit process.
1971 		 * The fields for the 64-bit quantities are not large enough.
1972 		 */
1973 		p = pnp->pr_common->prc_proc;
1974 		if (PROCESS_NOT_32BIT(p)) {
1975 			prunlock(pnp);
1976 			error = EOVERFLOW;
1977 		} else {
1978 			prgetstatus32(pnp->pr_common->prc_proc, sp,
1979 			    VTOZONE(PTOV(pnp)));
1980 			prunlock(pnp);
1981 			error = pr_uioread(sp, sizeof (*sp), uiop);
1982 		}
1983 	}
1984 	kmem_free((caddr_t)sp, sizeof (*sp));
1985 	return (error);
1986 }
1987 
1988 static int
1989 pr_read_lstatus_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
1990 {
1991 	proc_t *p;
1992 	kthread_t *t;
1993 	lwpdir_t *ldp;
1994 	size_t size;
1995 	prheader32_t *php;
1996 	lwpstatus32_t *sp;
1997 	int error;
1998 	int nlwp;
1999 	int i;
2000 
2001 	ASSERT(pnp->pr_type == PR_LSTATUS);
2002 
2003 	if ((error = prlock(pnp, ZNO)) != 0)
2004 		return (error);
2005 	p = pnp->pr_common->prc_proc;
2006 	/*
2007 	 * A 32-bit process cannot get the status of a 64-bit process.
2008 	 * The fields for the 64-bit quantities are not large enough.
2009 	 */
2010 	if (PROCESS_NOT_32BIT(p)) {
2011 		prunlock(pnp);
2012 		return (EOVERFLOW);
2013 	}
2014 	nlwp = p->p_lwpcnt;
2015 	size = sizeof (prheader32_t) + nlwp * LSPAN32(lwpstatus32_t);
2016 
2017 	/* drop p->p_lock to do kmem_alloc(KM_SLEEP) */
2018 	mutex_exit(&p->p_lock);
2019 	php = kmem_zalloc(size, KM_SLEEP);
2020 	mutex_enter(&p->p_lock);
2021 	/* p->p_lwpcnt can't change while process is locked */
2022 	ASSERT(nlwp == p->p_lwpcnt);
2023 
2024 	php->pr_nent = nlwp;
2025 	php->pr_entsize = LSPAN32(lwpstatus32_t);
2026 
2027 	sp = (lwpstatus32_t *)(php + 1);
2028 	for (ldp = p->p_lwpdir, i = 0; i < p->p_lwpdir_sz; i++, ldp++) {
2029 		if (ldp->ld_entry == NULL ||
2030 		    (t = ldp->ld_entry->le_thread) == NULL)
2031 			continue;
2032 		prgetlwpstatus32(t, sp, VTOZONE(PTOV(pnp)));
2033 		sp = (lwpstatus32_t *)((caddr_t)sp + LSPAN32(lwpstatus32_t));
2034 	}
2035 	prunlock(pnp);
2036 
2037 	error = pr_uioread(php, size, uiop);
2038 	kmem_free(php, size);
2039 	return (error);
2040 }
2041 
2042 static int
2043 pr_read_psinfo_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2044 {
2045 	psinfo32_t psinfo;
2046 	proc_t *p;
2047 	int error = 0;
2048 
2049 	ASSERT(pnp->pr_type == PR_PSINFO);
2050 
2051 	/*
2052 	 * We don't want the full treatment of prlock(pnp) here.
2053 	 * This file is world-readable and never goes invalid.
2054 	 * It doesn't matter if we are in the middle of an exec().
2055 	 */
2056 	p = pr_p_lock(pnp);
2057 	mutex_exit(&pr_pidlock);
2058 	if (p == NULL)
2059 		error = ENOENT;
2060 	else {
2061 		ASSERT(p == pnp->pr_common->prc_proc);
2062 		prgetpsinfo32(p, &psinfo);
2063 		prunlock(pnp);
2064 		error = pr_uioread(&psinfo, sizeof (psinfo), uiop);
2065 	}
2066 	return (error);
2067 }
2068 
2069 static int
2070 pr_read_lpsinfo_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2071 {
2072 	proc_t *p;
2073 	kthread_t *t;
2074 	lwpdir_t *ldp;
2075 	lwpent_t *lep;
2076 	size_t size;
2077 	prheader32_t *php;
2078 	lwpsinfo32_t *sp;
2079 	int error;
2080 	int nlwp;
2081 	int i;
2082 
2083 	ASSERT(pnp->pr_type == PR_LPSINFO);
2084 
2085 	/*
2086 	 * We don't want the full treatment of prlock(pnp) here.
2087 	 * This file is world-readable and never goes invalid.
2088 	 * It doesn't matter if we are in the middle of an exec().
2089 	 */
2090 	p = pr_p_lock(pnp);
2091 	mutex_exit(&pr_pidlock);
2092 	if (p == NULL)
2093 		return (ENOENT);
2094 	ASSERT(p == pnp->pr_common->prc_proc);
2095 	if ((nlwp = p->p_lwpcnt + p->p_zombcnt) == 0) {
2096 		prunlock(pnp);
2097 		return (ENOENT);
2098 	}
2099 	size = sizeof (prheader32_t) + nlwp * LSPAN32(lwpsinfo32_t);
2100 
2101 	/* drop p->p_lock to do kmem_alloc(KM_SLEEP) */
2102 	mutex_exit(&p->p_lock);
2103 	php = kmem_zalloc(size, KM_SLEEP);
2104 	mutex_enter(&p->p_lock);
2105 	/* p->p_lwpcnt can't change while process is locked */
2106 	ASSERT(nlwp == p->p_lwpcnt + p->p_zombcnt);
2107 
2108 	php->pr_nent = nlwp;
2109 	php->pr_entsize = LSPAN32(lwpsinfo32_t);
2110 
2111 	sp = (lwpsinfo32_t *)(php + 1);
2112 	for (ldp = p->p_lwpdir, i = 0; i < p->p_lwpdir_sz; i++, ldp++) {
2113 		if ((lep = ldp->ld_entry) == NULL)
2114 			continue;
2115 		if ((t = lep->le_thread) != NULL)
2116 			prgetlwpsinfo32(t, sp);
2117 		else {
2118 			bzero(sp, sizeof (*sp));
2119 			sp->pr_lwpid = lep->le_lwpid;
2120 			sp->pr_state = SZOMB;
2121 			sp->pr_sname = 'Z';
2122 			sp->pr_start.tv_sec = (time32_t)lep->le_start;
2123 		}
2124 		sp = (lwpsinfo32_t *)((caddr_t)sp + LSPAN32(lwpsinfo32_t));
2125 	}
2126 	prunlock(pnp);
2127 
2128 	error = pr_uioread(php, size, uiop);
2129 	kmem_free(php, size);
2130 	return (error);
2131 }
2132 
2133 static int
2134 pr_read_map_common_32(prnode_t *pnp, uio_t *uiop, prnodetype_t type)
2135 {
2136 	proc_t *p;
2137 	struct as *as;
2138 	list_t	iolhead;
2139 	int error;
2140 
2141 readmap32_common:
2142 	if ((error = prlock(pnp, ZNO)) != 0)
2143 		return (error);
2144 
2145 	p = pnp->pr_common->prc_proc;
2146 	as = p->p_as;
2147 
2148 	if ((p->p_flag & SSYS) || as == &kas) {
2149 		prunlock(pnp);
2150 		return (0);
2151 	}
2152 
2153 	if (PROCESS_NOT_32BIT(p)) {
2154 		prunlock(pnp);
2155 		return (EOVERFLOW);
2156 	}
2157 
2158 	if (!AS_LOCK_TRYENTER(as, RW_WRITER)) {
2159 		prunlock(pnp);
2160 		delay(1);
2161 		goto readmap32_common;
2162 	}
2163 	mutex_exit(&p->p_lock);
2164 
2165 	switch (type) {
2166 	case PR_XMAP:
2167 		error = prgetxmap32(p, &iolhead);
2168 		break;
2169 	case PR_RMAP:
2170 		error = prgetmap32(p, 1, &iolhead);
2171 		break;
2172 	case PR_MAP:
2173 		error = prgetmap32(p, 0, &iolhead);
2174 		break;
2175 	}
2176 	AS_LOCK_EXIT(as);
2177 	mutex_enter(&p->p_lock);
2178 	prunlock(pnp);
2179 
2180 	error = pr_iol_uiomove_and_free(&iolhead, uiop, error);
2181 
2182 	return (error);
2183 }
2184 
2185 static int
2186 pr_read_map_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2187 {
2188 	ASSERT(pnp->pr_type == PR_MAP);
2189 	return (pr_read_map_common_32(pnp, uiop, pnp->pr_type));
2190 }
2191 
2192 static int
2193 pr_read_rmap_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2194 {
2195 	ASSERT(pnp->pr_type == PR_RMAP);
2196 	return (pr_read_map_common_32(pnp, uiop, pnp->pr_type));
2197 }
2198 
2199 static int
2200 pr_read_xmap_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2201 {
2202 	ASSERT(pnp->pr_type == PR_XMAP);
2203 	return (pr_read_map_common_32(pnp, uiop, pnp->pr_type));
2204 }
2205 
2206 static int
2207 pr_read_sigact_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2208 {
2209 	int nsig = PROC_IS_BRANDED(curproc)? BROP(curproc)->b_nsig : NSIG;
2210 	proc_t *p;
2211 	struct sigaction32 *sap;
2212 	int sig;
2213 	int error;
2214 	user_t *up;
2215 
2216 	ASSERT(pnp->pr_type == PR_SIGACT);
2217 
2218 	/*
2219 	 * We kmem_alloc() the sigaction32 array because
2220 	 * it is so big it might blow the kernel stack.
2221 	 */
2222 	sap = kmem_alloc((nsig-1) * sizeof (struct sigaction32), KM_SLEEP);
2223 
2224 	if ((error = prlock(pnp, ZNO)) != 0)
2225 		goto out;
2226 	p = pnp->pr_common->prc_proc;
2227 
2228 	if (PROCESS_NOT_32BIT(p)) {
2229 		prunlock(pnp);
2230 		error = EOVERFLOW;
2231 		goto out;
2232 	}
2233 
2234 	if (uiop->uio_offset >= (nsig-1) * sizeof (struct sigaction32)) {
2235 		prunlock(pnp);
2236 		goto out;
2237 	}
2238 
2239 	up = PTOU(p);
2240 	for (sig = 1; sig < nsig; sig++)
2241 		prgetaction32(p, up, sig, &sap[sig-1]);
2242 	prunlock(pnp);
2243 
2244 	error = pr_uioread(sap, (nsig - 1) * sizeof (struct sigaction32), uiop);
2245 out:
2246 	kmem_free(sap, (nsig-1) * sizeof (struct sigaction32));
2247 	return (error);
2248 }
2249 
2250 static int
2251 pr_read_auxv_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2252 {
2253 	auxv32_t auxv[__KERN_NAUXV_IMPL];
2254 	proc_t *p;
2255 	user_t *up;
2256 	int error;
2257 	int i;
2258 
2259 	ASSERT(pnp->pr_type == PR_AUXV);
2260 
2261 	if ((error = prlock(pnp, ZNO)) != 0)
2262 		return (error);
2263 	p = pnp->pr_common->prc_proc;
2264 
2265 	if (PROCESS_NOT_32BIT(p)) {
2266 		prunlock(pnp);
2267 		return (EOVERFLOW);
2268 	}
2269 
2270 	if (uiop->uio_offset >= sizeof (auxv)) {
2271 		prunlock(pnp);
2272 		return (0);
2273 	}
2274 
2275 	up = PTOU(p);
2276 	for (i = 0; i < __KERN_NAUXV_IMPL; i++) {
2277 		auxv[i].a_type = (int32_t)up->u_auxv[i].a_type;
2278 		auxv[i].a_un.a_val = (int32_t)up->u_auxv[i].a_un.a_val;
2279 	}
2280 	prunlock(pnp);
2281 
2282 	return (pr_uioread(auxv, sizeof (auxv), uiop));
2283 }
2284 
2285 static int
2286 pr_read_usage_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2287 {
2288 	prhusage_t *pup;
2289 	prusage32_t *upup;
2290 	proc_t *p;
2291 	kthread_t *t;
2292 	int error;
2293 
2294 	ASSERT(pnp->pr_type == PR_USAGE);
2295 
2296 	/* allocate now, before locking the process */
2297 	pup = kmem_zalloc(sizeof (*pup), KM_SLEEP);
2298 	upup = kmem_alloc(sizeof (*upup), KM_SLEEP);
2299 
2300 	/*
2301 	 * We don't want the full treatment of prlock(pnp) here.
2302 	 * This file is world-readable and never goes invalid.
2303 	 * It doesn't matter if we are in the middle of an exec().
2304 	 */
2305 	p = pr_p_lock(pnp);
2306 	mutex_exit(&pr_pidlock);
2307 	if (p == NULL) {
2308 		error = ENOENT;
2309 		goto out;
2310 	}
2311 	ASSERT(p == pnp->pr_common->prc_proc);
2312 
2313 	if (uiop->uio_offset >= sizeof (prusage32_t)) {
2314 		prunlock(pnp);
2315 		error = 0;
2316 		goto out;
2317 	}
2318 
2319 	pup->pr_tstamp = gethrtime();
2320 
2321 	pup->pr_count  = p->p_defunct;
2322 	pup->pr_create = p->p_mstart;
2323 	pup->pr_term   = p->p_mterm;
2324 
2325 	pup->pr_rtime    = p->p_mlreal;
2326 	pup->pr_utime    = p->p_acct[LMS_USER];
2327 	pup->pr_stime    = p->p_acct[LMS_SYSTEM];
2328 	pup->pr_ttime    = p->p_acct[LMS_TRAP];
2329 	pup->pr_tftime   = p->p_acct[LMS_TFAULT];
2330 	pup->pr_dftime   = p->p_acct[LMS_DFAULT];
2331 	pup->pr_kftime   = p->p_acct[LMS_KFAULT];
2332 	pup->pr_ltime    = p->p_acct[LMS_USER_LOCK];
2333 	pup->pr_slptime  = p->p_acct[LMS_SLEEP];
2334 	pup->pr_wtime    = p->p_acct[LMS_WAIT_CPU];
2335 	pup->pr_stoptime = p->p_acct[LMS_STOPPED];
2336 
2337 	pup->pr_minf  = p->p_ru.minflt;
2338 	pup->pr_majf  = p->p_ru.majflt;
2339 	pup->pr_nswap = p->p_ru.nswap;
2340 	pup->pr_inblk = p->p_ru.inblock;
2341 	pup->pr_oublk = p->p_ru.oublock;
2342 	pup->pr_msnd  = p->p_ru.msgsnd;
2343 	pup->pr_mrcv  = p->p_ru.msgrcv;
2344 	pup->pr_sigs  = p->p_ru.nsignals;
2345 	pup->pr_vctx  = p->p_ru.nvcsw;
2346 	pup->pr_ictx  = p->p_ru.nivcsw;
2347 	pup->pr_sysc  = p->p_ru.sysc;
2348 	pup->pr_ioch  = p->p_ru.ioch;
2349 
2350 	/*
2351 	 * Add the usage information for each active lwp.
2352 	 */
2353 	if ((t = p->p_tlist) != NULL &&
2354 	    !(pnp->pr_pcommon->prc_flags & PRC_DESTROY)) {
2355 		do {
2356 			if (t->t_proc_flag & TP_LWPEXIT)
2357 				continue;
2358 			pup->pr_count++;
2359 			praddusage(t, pup);
2360 		} while ((t = t->t_forw) != p->p_tlist);
2361 	}
2362 
2363 	prunlock(pnp);
2364 
2365 	prcvtusage32(pup, upup);
2366 
2367 	error = pr_uioread(upup, sizeof (prusage32_t), uiop);
2368 out:
2369 	kmem_free(pup, sizeof (*pup));
2370 	kmem_free(upup, sizeof (*upup));
2371 	return (error);
2372 }
2373 
2374 static int
2375 pr_read_lusage_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2376 {
2377 	int nlwp;
2378 	prhusage_t *pup;
2379 	prheader32_t *php;
2380 	prusage32_t *upup;
2381 	size_t size;
2382 	hrtime_t curtime;
2383 	proc_t *p;
2384 	kthread_t *t;
2385 	lwpdir_t *ldp;
2386 	int error;
2387 	int i;
2388 
2389 	ASSERT(pnp->pr_type == PR_LUSAGE);
2390 
2391 	/*
2392 	 * We don't want the full treatment of prlock(pnp) here.
2393 	 * This file is world-readable and never goes invalid.
2394 	 * It doesn't matter if we are in the middle of an exec().
2395 	 */
2396 	p = pr_p_lock(pnp);
2397 	mutex_exit(&pr_pidlock);
2398 	if (p == NULL)
2399 		return (ENOENT);
2400 	ASSERT(p == pnp->pr_common->prc_proc);
2401 	if ((nlwp = p->p_lwpcnt) == 0) {
2402 		prunlock(pnp);
2403 		return (ENOENT);
2404 	}
2405 
2406 	size = sizeof (prheader32_t) + (nlwp + 1) * LSPAN32(prusage32_t);
2407 	if (uiop->uio_offset >= size) {
2408 		prunlock(pnp);
2409 		return (0);
2410 	}
2411 
2412 	/* drop p->p_lock to do kmem_alloc(KM_SLEEP) */
2413 	mutex_exit(&p->p_lock);
2414 	pup = kmem_zalloc(size + sizeof (prhusage_t), KM_SLEEP);
2415 	mutex_enter(&p->p_lock);
2416 	/* p->p_lwpcnt can't change while process is locked */
2417 	ASSERT(nlwp == p->p_lwpcnt);
2418 
2419 	php = (prheader32_t *)(pup + 1);
2420 	upup = (prusage32_t *)(php + 1);
2421 
2422 	php->pr_nent = nlwp + 1;
2423 	php->pr_entsize = LSPAN32(prusage32_t);
2424 
2425 	curtime = gethrtime();
2426 
2427 	/*
2428 	 * First the summation over defunct lwps.
2429 	 */
2430 	pup->pr_count  = p->p_defunct;
2431 	pup->pr_tstamp = curtime;
2432 	pup->pr_create = p->p_mstart;
2433 	pup->pr_term   = p->p_mterm;
2434 
2435 	pup->pr_rtime    = p->p_mlreal;
2436 	pup->pr_utime    = p->p_acct[LMS_USER];
2437 	pup->pr_stime    = p->p_acct[LMS_SYSTEM];
2438 	pup->pr_ttime    = p->p_acct[LMS_TRAP];
2439 	pup->pr_tftime   = p->p_acct[LMS_TFAULT];
2440 	pup->pr_dftime   = p->p_acct[LMS_DFAULT];
2441 	pup->pr_kftime   = p->p_acct[LMS_KFAULT];
2442 	pup->pr_ltime    = p->p_acct[LMS_USER_LOCK];
2443 	pup->pr_slptime  = p->p_acct[LMS_SLEEP];
2444 	pup->pr_wtime    = p->p_acct[LMS_WAIT_CPU];
2445 	pup->pr_stoptime = p->p_acct[LMS_STOPPED];
2446 
2447 	pup->pr_minf  = p->p_ru.minflt;
2448 	pup->pr_majf  = p->p_ru.majflt;
2449 	pup->pr_nswap = p->p_ru.nswap;
2450 	pup->pr_inblk = p->p_ru.inblock;
2451 	pup->pr_oublk = p->p_ru.oublock;
2452 	pup->pr_msnd  = p->p_ru.msgsnd;
2453 	pup->pr_mrcv  = p->p_ru.msgrcv;
2454 	pup->pr_sigs  = p->p_ru.nsignals;
2455 	pup->pr_vctx  = p->p_ru.nvcsw;
2456 	pup->pr_ictx  = p->p_ru.nivcsw;
2457 	pup->pr_sysc  = p->p_ru.sysc;
2458 	pup->pr_ioch  = p->p_ru.ioch;
2459 
2460 	prcvtusage32(pup, upup);
2461 
2462 	/*
2463 	 * Fill one prusage struct for each active lwp.
2464 	 */
2465 	for (ldp = p->p_lwpdir, i = 0; i < p->p_lwpdir_sz; i++, ldp++) {
2466 		if (ldp->ld_entry == NULL ||
2467 		    (t = ldp->ld_entry->le_thread) == NULL)
2468 			continue;
2469 		ASSERT(!(t->t_proc_flag & TP_LWPEXIT));
2470 		ASSERT(nlwp > 0);
2471 		--nlwp;
2472 		upup = (prusage32_t *)
2473 		    ((caddr_t)upup + LSPAN32(prusage32_t));
2474 		prgetusage(t, pup);
2475 		prcvtusage32(pup, upup);
2476 	}
2477 	ASSERT(nlwp == 0);
2478 
2479 	prunlock(pnp);
2480 
2481 	error = pr_uioread(php, size, uiop);
2482 	kmem_free(pup, size + sizeof (prhusage_t));
2483 	return (error);
2484 }
2485 
2486 static int
2487 pr_read_pagedata_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2488 {
2489 	proc_t *p;
2490 	int error;
2491 
2492 	ASSERT(pnp->pr_type == PR_PAGEDATA);
2493 
2494 	if ((error = prlock(pnp, ZNO)) != 0)
2495 		return (error);
2496 
2497 	p = pnp->pr_common->prc_proc;
2498 	if ((p->p_flag & SSYS) || p->p_as == &kas) {
2499 		prunlock(pnp);
2500 		return (0);
2501 	}
2502 
2503 	if (PROCESS_NOT_32BIT(p)) {
2504 		prunlock(pnp);
2505 		return (EOVERFLOW);
2506 	}
2507 
2508 	mutex_exit(&p->p_lock);
2509 	error = prpdread32(p, pnp->pr_hatid, uiop);
2510 	mutex_enter(&p->p_lock);
2511 
2512 	prunlock(pnp);
2513 	return (error);
2514 }
2515 
2516 static int
2517 pr_read_opagedata_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2518 {
2519 	proc_t *p;
2520 	struct as *as;
2521 	int error;
2522 
2523 	ASSERT(pnp->pr_type == PR_OPAGEDATA);
2524 
2525 	if ((error = prlock(pnp, ZNO)) != 0)
2526 		return (error);
2527 
2528 	p = pnp->pr_common->prc_proc;
2529 	as = p->p_as;
2530 
2531 	if ((p->p_flag & SSYS) || as == &kas) {
2532 		prunlock(pnp);
2533 		return (0);
2534 	}
2535 
2536 	if (PROCESS_NOT_32BIT(p)) {
2537 		prunlock(pnp);
2538 		return (EOVERFLOW);
2539 	}
2540 
2541 	mutex_exit(&p->p_lock);
2542 	error = oprpdread32(as, pnp->pr_hatid, uiop);
2543 	mutex_enter(&p->p_lock);
2544 
2545 	prunlock(pnp);
2546 	return (error);
2547 }
2548 
2549 static int
2550 pr_read_watch_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2551 {
2552 	proc_t *p;
2553 	int error;
2554 	prwatch32_t *Bpwp;
2555 	size_t size;
2556 	prwatch32_t *pwp;
2557 	int nwarea;
2558 	struct watched_area *pwarea;
2559 
2560 	ASSERT(pnp->pr_type == PR_WATCH);
2561 
2562 	if ((error = prlock(pnp, ZNO)) != 0)
2563 		return (error);
2564 
2565 	p = pnp->pr_common->prc_proc;
2566 	if (PROCESS_NOT_32BIT(p)) {
2567 		prunlock(pnp);
2568 		return (EOVERFLOW);
2569 	}
2570 	nwarea = avl_numnodes(&p->p_warea);
2571 	size = nwarea * sizeof (prwatch32_t);
2572 	if (uiop->uio_offset >= size) {
2573 		prunlock(pnp);
2574 		return (0);
2575 	}
2576 
2577 	/* drop p->p_lock to do kmem_alloc(KM_SLEEP) */
2578 	mutex_exit(&p->p_lock);
2579 	Bpwp = pwp = kmem_zalloc(size, KM_SLEEP);
2580 	mutex_enter(&p->p_lock);
2581 	/* p->p_nwarea can't change while process is locked */
2582 	ASSERT(nwarea == avl_numnodes(&p->p_warea));
2583 
2584 	/* gather the watched areas */
2585 	for (pwarea = avl_first(&p->p_warea); pwarea != NULL;
2586 	    pwarea = AVL_NEXT(&p->p_warea, pwarea), pwp++) {
2587 		pwp->pr_vaddr = (caddr32_t)(uintptr_t)pwarea->wa_vaddr;
2588 		pwp->pr_size = (size32_t)(pwarea->wa_eaddr - pwarea->wa_vaddr);
2589 		pwp->pr_wflags = (int)pwarea->wa_flags;
2590 	}
2591 
2592 	prunlock(pnp);
2593 
2594 	error = pr_uioread(Bpwp, size, uiop);
2595 	kmem_free(Bpwp, size);
2596 	return (error);
2597 }
2598 
2599 static int
2600 pr_read_lwpstatus_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2601 {
2602 	lwpstatus32_t *sp;
2603 	proc_t *p;
2604 	int error;
2605 
2606 	ASSERT(pnp->pr_type == PR_LWPSTATUS);
2607 
2608 	/*
2609 	 * We kmem_alloc() the lwpstatus structure because
2610 	 * it is so big it might blow the kernel stack.
2611 	 */
2612 	sp = kmem_alloc(sizeof (*sp), KM_SLEEP);
2613 
2614 	if ((error = prlock(pnp, ZNO)) != 0)
2615 		goto out;
2616 
2617 	/*
2618 	 * A 32-bit process cannot get the status of a 64-bit process.
2619 	 * The fields for the 64-bit quantities are not large enough.
2620 	 */
2621 	p = pnp->pr_common->prc_proc;
2622 	if (PROCESS_NOT_32BIT(p)) {
2623 		prunlock(pnp);
2624 		error = EOVERFLOW;
2625 		goto out;
2626 	}
2627 
2628 	if (uiop->uio_offset >= sizeof (*sp)) {
2629 		prunlock(pnp);
2630 		goto out;
2631 	}
2632 
2633 	prgetlwpstatus32(pnp->pr_common->prc_thread, sp, VTOZONE(PTOV(pnp)));
2634 	prunlock(pnp);
2635 
2636 	error = pr_uioread(sp, sizeof (*sp), uiop);
2637 out:
2638 	kmem_free(sp, sizeof (*sp));
2639 	return (error);
2640 }
2641 
2642 static int
2643 pr_read_lwpsinfo_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2644 {
2645 	lwpsinfo32_t lwpsinfo;
2646 	proc_t *p;
2647 	kthread_t *t;
2648 	lwpent_t *lep;
2649 
2650 	ASSERT(pnp->pr_type == PR_LWPSINFO);
2651 
2652 	/*
2653 	 * We don't want the full treatment of prlock(pnp) here.
2654 	 * This file is world-readable and never goes invalid.
2655 	 * It doesn't matter if we are in the middle of an exec().
2656 	 */
2657 	p = pr_p_lock(pnp);
2658 	mutex_exit(&pr_pidlock);
2659 	if (p == NULL)
2660 		return (ENOENT);
2661 	ASSERT(p == pnp->pr_common->prc_proc);
2662 	if (pnp->pr_common->prc_tslot == -1) {
2663 		prunlock(pnp);
2664 		return (ENOENT);
2665 	}
2666 
2667 	if (uiop->uio_offset >= sizeof (lwpsinfo)) {
2668 		prunlock(pnp);
2669 		return (0);
2670 	}
2671 
2672 	if ((t = pnp->pr_common->prc_thread) != NULL)
2673 		prgetlwpsinfo32(t, &lwpsinfo);
2674 	else {
2675 		lep = p->p_lwpdir[pnp->pr_common->prc_tslot].ld_entry;
2676 		bzero(&lwpsinfo, sizeof (lwpsinfo));
2677 		lwpsinfo.pr_lwpid = lep->le_lwpid;
2678 		lwpsinfo.pr_state = SZOMB;
2679 		lwpsinfo.pr_sname = 'Z';
2680 		lwpsinfo.pr_start.tv_sec = (time32_t)lep->le_start;
2681 	}
2682 	prunlock(pnp);
2683 
2684 	return (pr_uioread(&lwpsinfo, sizeof (lwpsinfo), uiop));
2685 }
2686 
2687 static int
2688 pr_read_lwpusage_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2689 {
2690 	prhusage_t *pup;
2691 	prusage32_t *upup;
2692 	proc_t *p;
2693 	int error;
2694 
2695 	ASSERT(pnp->pr_type == PR_LWPUSAGE);
2696 
2697 	/* allocate now, before locking the process */
2698 	pup = kmem_zalloc(sizeof (*pup), KM_SLEEP);
2699 	upup = kmem_alloc(sizeof (*upup), KM_SLEEP);
2700 
2701 	/*
2702 	 * We don't want the full treatment of prlock(pnp) here.
2703 	 * This file is world-readable and never goes invalid.
2704 	 * It doesn't matter if we are in the middle of an exec().
2705 	 */
2706 	p = pr_p_lock(pnp);
2707 	mutex_exit(&pr_pidlock);
2708 	if (p == NULL) {
2709 		error = ENOENT;
2710 		goto out;
2711 	}
2712 	ASSERT(p == pnp->pr_common->prc_proc);
2713 	if (pnp->pr_common->prc_thread == NULL) {
2714 		prunlock(pnp);
2715 		error = ENOENT;
2716 		goto out;
2717 	}
2718 	if (uiop->uio_offset >= sizeof (prusage32_t)) {
2719 		prunlock(pnp);
2720 		error = 0;
2721 		goto out;
2722 	}
2723 
2724 	pup->pr_tstamp = gethrtime();
2725 	prgetusage(pnp->pr_common->prc_thread, pup);
2726 
2727 	prunlock(pnp);
2728 
2729 	prcvtusage32(pup, upup);
2730 
2731 	error = pr_uioread(upup, sizeof (prusage32_t), uiop);
2732 out:
2733 	kmem_free(pup, sizeof (*pup));
2734 	kmem_free(upup, sizeof (*upup));
2735 	return (error);
2736 }
2737 
2738 static int
2739 pr_read_spymaster_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2740 {
2741 	psinfo32_t psinfo;
2742 	int error;
2743 	klwp_t *lwp;
2744 
2745 	ASSERT(pnp->pr_type == PR_SPYMASTER);
2746 
2747 	if ((error = prlock(pnp, ZNO)) != 0)
2748 		return (error);
2749 
2750 	if (pnp->pr_common->prc_thread == NULL) {
2751 		prunlock(pnp);
2752 		return (0);
2753 	}
2754 
2755 	lwp = pnp->pr_common->prc_thread->t_lwp;
2756 
2757 	if (lwp->lwp_spymaster == NULL) {
2758 		prunlock(pnp);
2759 		return (0);
2760 	}
2761 
2762 	psinfo_kto32(lwp->lwp_spymaster, &psinfo);
2763 	prunlock(pnp);
2764 
2765 	return (pr_uioread(&psinfo, sizeof (psinfo), uiop));
2766 }
2767 
2768 #if defined(__sparc)
2769 static int
2770 pr_read_gwindows_32(prnode_t *pnp, uio_t *uiop, cred_t *cr)
2771 {
2772 	proc_t *p;
2773 	kthread_t *t;
2774 	gwindows32_t *gwp;
2775 	int error;
2776 	size_t size;
2777 
2778 	ASSERT(pnp->pr_type == PR_GWINDOWS);
2779 
2780 	gwp = kmem_zalloc(sizeof (gwindows32_t), KM_SLEEP);
2781 
2782 	if ((error = prlock(pnp, ZNO)) != 0)
2783 		goto out;
2784 
2785 	p = pnp->pr_common->prc_proc;
2786 	t = pnp->pr_common->prc_thread;
2787 
2788 	if (PROCESS_NOT_32BIT(p)) {
2789 		prunlock(pnp);
2790 		error = EOVERFLOW;
2791 		goto out;
2792 	}
2793 
2794 	/*
2795 	 * Drop p->p_lock while touching the stack.
2796 	 * The P_PR_LOCK flag prevents the lwp from
2797 	 * disappearing while we do this.
2798 	 */
2799 	mutex_exit(&p->p_lock);
2800 	if ((size = prnwindows(ttolwp(t))) != 0)
2801 		size = sizeof (gwindows32_t) -
2802 		    (SPARC_MAXREGWINDOW - size) * sizeof (struct rwindow32);
2803 	if (uiop->uio_offset >= size) {
2804 		mutex_enter(&p->p_lock);
2805 		prunlock(pnp);
2806 		goto out;
2807 	}
2808 	prgetwindows32(ttolwp(t), gwp);
2809 	mutex_enter(&p->p_lock);
2810 	prunlock(pnp);
2811 
2812 	error = pr_uioread(gwp, size, uiop);
2813 out:
2814 	kmem_free(gwp, sizeof (gwindows32_t));
2815 	return (error);
2816 }
2817 #endif	/* __sparc */
2818 
2819 #endif	/* _SYSCALL32_IMPL */
2820 
2821 /* ARGSUSED */
2822 static int
2823 prread(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr, caller_context_t *ct)
2824 {
2825 	prnode_t *pnp = VTOP(vp);
2826 
2827 	ASSERT(pnp->pr_type < PR_NFILES);
2828 
2829 #ifdef _SYSCALL32_IMPL
2830 	/*
2831 	 * What is read from the /proc files depends on the data
2832 	 * model of the caller.  An LP64 process will see LP64
2833 	 * data.  An ILP32 process will see ILP32 data.
2834 	 */
2835 	if (curproc->p_model == DATAMODEL_LP64)
2836 		return (pr_read_function[pnp->pr_type](pnp, uiop, cr));
2837 	else
2838 		return (pr_read_function_32[pnp->pr_type](pnp, uiop, cr));
2839 #else
2840 	return (pr_read_function[pnp->pr_type](pnp, uiop, cr));
2841 #endif
2842 }
2843 
2844 /* Note we intentionally don't handle partial writes/updates. */
2845 static int
2846 pr_write_lwpname(prnode_t *pnp, uio_t *uiop)
2847 {
2848 	kthread_t *t = NULL;
2849 	char *lwpname;
2850 	int error;
2851 
2852 	lwpname = kmem_zalloc(THREAD_NAME_MAX, KM_SLEEP);
2853 
2854 	if ((error = uiomove(lwpname, THREAD_NAME_MAX, UIO_WRITE, uiop)) != 0) {
2855 		kmem_free(lwpname, THREAD_NAME_MAX);
2856 		return (error);
2857 	}
2858 
2859 	/* Somebody tried to write too long a thread name... */
2860 	if (lwpname[THREAD_NAME_MAX - 1] != '\0' || uiop->uio_resid > 0) {
2861 		kmem_free(lwpname, THREAD_NAME_MAX);
2862 		return (EIO);
2863 	}
2864 
2865 	VERIFY3U(lwpname[THREAD_NAME_MAX - 1], ==, '\0');
2866 
2867 	for (size_t i = 0; lwpname[i] != '\0'; i++) {
2868 		if (!ISPRINT(lwpname[i])) {
2869 			kmem_free(lwpname, THREAD_NAME_MAX);
2870 			return (EINVAL);
2871 		}
2872 	}
2873 
2874 	/* Equivalent of thread_setname(), but with the ZNO magic. */
2875 	if ((error = prlock(pnp, ZNO)) != 0) {
2876 		kmem_free(lwpname, THREAD_NAME_MAX);
2877 		return (error);
2878 	}
2879 
2880 	t = pnp->pr_common->prc_thread;
2881 	if (t->t_name == NULL) {
2882 		t->t_name = lwpname;
2883 	} else {
2884 		(void) strlcpy(t->t_name, lwpname, THREAD_NAME_MAX);
2885 		kmem_free(lwpname, THREAD_NAME_MAX);
2886 	}
2887 
2888 	prunlock(pnp);
2889 	return (0);
2890 }
2891 
2892 /* ARGSUSED */
2893 static int
2894 prwrite(vnode_t *vp, uio_t *uiop, int ioflag, cred_t *cr, caller_context_t *ct)
2895 {
2896 	prnode_t *pnp = VTOP(vp);
2897 	int old = 0;
2898 	int error;
2899 	ssize_t resid;
2900 
2901 	ASSERT(pnp->pr_type < PR_NFILES);
2902 
2903 	/*
2904 	 * Only a handful of /proc files are writable, enumerate them here.
2905 	 */
2906 	switch (pnp->pr_type) {
2907 	case PR_PIDDIR:		/* directory write()s: visceral revulsion. */
2908 		ASSERT(pnp->pr_pidfile != NULL);
2909 		/* use the underlying PR_PIDFILE to write the process */
2910 		vp = pnp->pr_pidfile;
2911 		pnp = VTOP(vp);
2912 		ASSERT(pnp->pr_type == PR_PIDFILE);
2913 		/* FALLTHROUGH */
2914 	case PR_PIDFILE:
2915 	case PR_LWPIDFILE:
2916 		old = 1;
2917 		/* FALLTHROUGH */
2918 	case PR_AS:
2919 		if ((error = prlock(pnp, ZNO)) == 0) {
2920 			proc_t *p = pnp->pr_common->prc_proc;
2921 			struct as *as = p->p_as;
2922 
2923 			if ((p->p_flag & SSYS) || as == &kas) {
2924 				/*
2925 				 * /proc I/O cannot be done to a system process.
2926 				 */
2927 				error = EIO;
2928 #ifdef _SYSCALL32_IMPL
2929 			} else if (curproc->p_model == DATAMODEL_ILP32 &&
2930 			    PROCESS_NOT_32BIT(p)) {
2931 				error = EOVERFLOW;
2932 #endif
2933 			} else {
2934 				/*
2935 				 * See comments above (pr_read_pidfile)
2936 				 * about this locking dance.
2937 				 */
2938 				mutex_exit(&p->p_lock);
2939 				error = prusrio(p, UIO_WRITE, uiop, old);
2940 				mutex_enter(&p->p_lock);
2941 			}
2942 			prunlock(pnp);
2943 		}
2944 		return (error);
2945 
2946 	case PR_CTL:
2947 	case PR_LWPCTL:
2948 		resid = uiop->uio_resid;
2949 		/*
2950 		 * Perform the action on the control file
2951 		 * by passing curthreads credentials
2952 		 * and not target process's credentials.
2953 		 */
2954 #ifdef _SYSCALL32_IMPL
2955 		if (curproc->p_model == DATAMODEL_ILP32)
2956 			error = prwritectl32(vp, uiop, CRED());
2957 		else
2958 			error = prwritectl(vp, uiop, CRED());
2959 #else
2960 		error = prwritectl(vp, uiop, CRED());
2961 #endif
2962 		/*
2963 		 * This hack makes sure that the EINTR is passed
2964 		 * all the way back to the caller's write() call.
2965 		 */
2966 		if (error == EINTR)
2967 			uiop->uio_resid = resid;
2968 		return (error);
2969 
2970 	case PR_LWPNAME:
2971 		return (pr_write_lwpname(pnp, uiop));
2972 
2973 	default:
2974 		return ((vp->v_type == VDIR)? EISDIR : EBADF);
2975 	}
2976 	/* NOTREACHED */
2977 }
2978 
2979 static int
2980 prgetattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2981     caller_context_t *ct)
2982 {
2983 	prnode_t *pnp = VTOP(vp);
2984 	prnodetype_t type = pnp->pr_type;
2985 	prcommon_t *pcp;
2986 	proc_t *p;
2987 	struct as *as;
2988 	int error;
2989 	vnode_t *rvp;
2990 	timestruc_t now;
2991 	extern uint_t nproc;
2992 	int ngroups;
2993 	int nsig;
2994 
2995 	/*
2996 	 * This ugly bit of code allows us to keep both versions of this
2997 	 * function from the same source.
2998 	 */
2999 #ifdef _LP64
3000 	int iam32bit = (curproc->p_model == DATAMODEL_ILP32);
3001 #define	PR_OBJSIZE(obj32, obj64)	\
3002 	(iam32bit ? sizeof (obj32) : sizeof (obj64))
3003 #define	PR_OBJSPAN(obj32, obj64)	\
3004 	(iam32bit ? LSPAN32(obj32) : LSPAN(obj64))
3005 #else
3006 #define	PR_OBJSIZE(obj32, obj64)	\
3007 	(sizeof (obj64))
3008 #define	PR_OBJSPAN(obj32, obj64)	\
3009 	(LSPAN(obj64))
3010 #endif
3011 
3012 	/*
3013 	 * Return all the attributes.  Should be refined
3014 	 * so that it returns only those asked for.
3015 	 * Most of this is complete fakery anyway.
3016 	 */
3017 
3018 	/*
3019 	 * For files in the /proc/<pid>/object directory,
3020 	 * return the attributes of the underlying object.
3021 	 * For files in the /proc/<pid>/fd directory,
3022 	 * return the attributes of the underlying file, but
3023 	 * make it look inaccessible if it is not a regular file.
3024 	 * Make directories look like symlinks.
3025 	 */
3026 	switch (type) {
3027 	case PR_CURDIR:
3028 	case PR_ROOTDIR:
3029 		if (!(flags & ATTR_REAL))
3030 			break;
3031 		/* restrict full knowledge of the attributes to owner or root */
3032 		if ((error = praccess(vp, 0, 0, cr, ct)) != 0)
3033 			return (error);
3034 		/* FALLTHROUGH */
3035 	case PR_OBJECT:
3036 	case PR_FD:
3037 		rvp = pnp->pr_realvp;
3038 		error = VOP_GETATTR(rvp, vap, flags, cr, ct);
3039 		if (error)
3040 			return (error);
3041 		if (type == PR_FD) {
3042 			if (rvp->v_type != VREG && rvp->v_type != VDIR)
3043 				vap->va_mode = 0;
3044 			else
3045 				vap->va_mode &= pnp->pr_mode;
3046 		}
3047 		if (type == PR_OBJECT)
3048 			vap->va_mode &= 07555;
3049 		if (rvp->v_type == VDIR && !(flags & ATTR_REAL)) {
3050 			vap->va_type = VLNK;
3051 			vap->va_size = 0;
3052 			vap->va_nlink = 1;
3053 		}
3054 		return (0);
3055 	default:
3056 		break;
3057 	}
3058 
3059 	bzero(vap, sizeof (*vap));
3060 	/*
3061 	 * Large Files: Internally proc now uses VPROC to indicate
3062 	 * a proc file. Since we have been returning VREG through
3063 	 * VOP_GETATTR() until now, we continue to do this so as
3064 	 * not to break apps depending on this return value.
3065 	 */
3066 	vap->va_type = (vp->v_type == VPROC) ? VREG : vp->v_type;
3067 	vap->va_mode = pnp->pr_mode;
3068 	vap->va_fsid = vp->v_vfsp->vfs_dev;
3069 	vap->va_blksize = DEV_BSIZE;
3070 	vap->va_rdev = 0;
3071 	vap->va_seq = 0;
3072 
3073 	if (type == PR_PROCDIR) {
3074 		vap->va_uid = 0;
3075 		vap->va_gid = 0;
3076 		vap->va_nlink = nproc + 2;
3077 		vap->va_nodeid = (ino64_t)PRROOTINO;
3078 		gethrestime(&now);
3079 		vap->va_atime = vap->va_mtime = vap->va_ctime = now;
3080 		vap->va_size = (v.v_proc + 2) * PRSDSIZE;
3081 		vap->va_nblocks = btod(vap->va_size);
3082 		return (0);
3083 	}
3084 
3085 	/*
3086 	 * /proc/<pid>/self is a symbolic link, and has no prcommon member
3087 	 */
3088 	if (type == PR_SELF) {
3089 		vap->va_uid = crgetruid(CRED());
3090 		vap->va_gid = crgetrgid(CRED());
3091 		vap->va_nodeid = (ino64_t)PR_SELF;
3092 		gethrestime(&now);
3093 		vap->va_atime = vap->va_mtime = vap->va_ctime = now;
3094 		vap->va_nlink = 1;
3095 		vap->va_type = VLNK;
3096 		vap->va_size = 0;
3097 		return (0);
3098 	}
3099 
3100 	/* A subset of prlock(pnp...) */
3101 	p = pr_p_lock(pnp);
3102 	mutex_exit(&pr_pidlock);
3103 	if (p == NULL)
3104 		return (ENOENT);
3105 	pcp = pnp->pr_common;
3106 
3107 	/*
3108 	 * Because we're performing a subset of prlock() inline here, we must
3109 	 * follow prlock's semantics when encountering a zombie process
3110 	 * (PRC_DESTROY flag is set) or an exiting process (SEXITING flag is
3111 	 * set). Those semantics indicate acting as if the process is no
3112 	 * longer there (return ENOENT).
3113 	 *
3114 	 * If we chose to proceed here regardless, we may encounter issues
3115 	 * when we drop the p_lock (see PR_OBJECTDIR, PR_PATHDIR, PR_*MAP,
3116 	 * PR_LDT, and PR_*PAGEDATA below). A process-cleanup which was
3117 	 * blocked on p_lock may ignore the P_PR_LOCK flag we set above, since
3118 	 * it set one of PRC_DESTROY or SEXITING. If the process then gets
3119 	 * destroyed our "p" will be useless, as will its p_lock.
3120 	 *
3121 	 * It may be desirable to move this check to only places further down
3122 	 * prior to actual droppages of p->p_lock, but for now, we're playing
3123 	 * it safe and checking here immediately, like prlock() does..
3124 	 */
3125 	if (((pcp->prc_flags & PRC_DESTROY) || (p->p_flag & SEXITING))) {
3126 		prunlock(pnp);
3127 		return (ENOENT);
3128 	}
3129 
3130 	mutex_enter(&p->p_crlock);
3131 	vap->va_uid = crgetruid(p->p_cred);
3132 	vap->va_gid = crgetrgid(p->p_cred);
3133 	mutex_exit(&p->p_crlock);
3134 
3135 	vap->va_nlink = 1;
3136 	vap->va_nodeid = pnp->pr_ino? pnp->pr_ino :
3137 	    pmkino(pcp->prc_tslot, pcp->prc_slot, pnp->pr_type);
3138 	if ((pcp->prc_flags & PRC_LWP) && pcp->prc_tslot != -1) {
3139 		vap->va_atime.tv_sec = vap->va_mtime.tv_sec =
3140 		    vap->va_ctime.tv_sec =
3141 		    p->p_lwpdir[pcp->prc_tslot].ld_entry->le_start;
3142 		vap->va_atime.tv_nsec = vap->va_mtime.tv_nsec =
3143 		    vap->va_ctime.tv_nsec = 0;
3144 	} else {
3145 		user_t *up = PTOU(p);
3146 		vap->va_atime.tv_sec = vap->va_mtime.tv_sec =
3147 		    vap->va_ctime.tv_sec = up->u_start.tv_sec;
3148 		vap->va_atime.tv_nsec = vap->va_mtime.tv_nsec =
3149 		    vap->va_ctime.tv_nsec = up->u_start.tv_nsec;
3150 	}
3151 
3152 	switch (type) {
3153 	case PR_PIDDIR:
3154 		/* va_nlink: count 'lwp', 'object' and 'fd' directory links */
3155 		vap->va_nlink = 5;
3156 		vap->va_size = sizeof (piddir);
3157 		break;
3158 	case PR_OBJECTDIR:
3159 		if ((p->p_flag & SSYS) || (as = p->p_as) == &kas)
3160 			vap->va_size = 2 * PRSDSIZE;
3161 		else {
3162 			mutex_exit(&p->p_lock);
3163 			AS_LOCK_ENTER(as, RW_WRITER);
3164 			if (as->a_updatedir)
3165 				rebuild_objdir(as);
3166 			vap->va_size = (as->a_sizedir + 2) * PRSDSIZE;
3167 			AS_LOCK_EXIT(as);
3168 			mutex_enter(&p->p_lock);
3169 		}
3170 		vap->va_nlink = 2;
3171 		break;
3172 	case PR_PATHDIR:
3173 		if ((p->p_flag & SSYS) || (as = p->p_as) == &kas)
3174 			vap->va_size = (P_FINFO(p)->fi_nfiles + 4) * PRSDSIZE;
3175 		else {
3176 			mutex_exit(&p->p_lock);
3177 			AS_LOCK_ENTER(as, RW_WRITER);
3178 			if (as->a_updatedir)
3179 				rebuild_objdir(as);
3180 			vap->va_size = (as->a_sizedir + 4 +
3181 			    P_FINFO(p)->fi_nfiles) * PRSDSIZE;
3182 			AS_LOCK_EXIT(as);
3183 			mutex_enter(&p->p_lock);
3184 		}
3185 		vap->va_nlink = 2;
3186 		break;
3187 	case PR_PATH:
3188 	case PR_CURDIR:
3189 	case PR_ROOTDIR:
3190 	case PR_CT:
3191 		vap->va_type = VLNK;
3192 		vap->va_size = 0;
3193 		break;
3194 	case PR_FDDIR:
3195 	case PR_FDINFODIR:
3196 		vap->va_nlink = 2;
3197 		vap->va_size = (P_FINFO(p)->fi_nfiles + 2) * PRSDSIZE;
3198 		break;
3199 	case PR_FDINFO: {
3200 		file_t *fp;
3201 		int fd = pnp->pr_index;
3202 
3203 		fp = pr_getf(p, fd, NULL);
3204 		if (fp == NULL) {
3205 			prunlock(pnp);
3206 			return (ENOENT);
3207 		}
3208 		prunlock(pnp);
3209 		vap->va_size = prgetfdinfosize(p, fp->f_vnode, cr);
3210 		vap->va_nblocks = (fsblkcnt64_t)btod(vap->va_size);
3211 		pr_releasef(fp);
3212 		return (0);
3213 	}
3214 	case PR_LWPDIR:
3215 		/*
3216 		 * va_nlink: count each lwp as a directory link.
3217 		 * va_size: size of p_lwpdir + 2
3218 		 */
3219 		vap->va_nlink = p->p_lwpcnt + p->p_zombcnt + 2;
3220 		vap->va_size = (p->p_lwpdir_sz + 2) * PRSDSIZE;
3221 		break;
3222 	case PR_LWPIDDIR:
3223 		vap->va_nlink = 2;
3224 		vap->va_size = sizeof (lwpiddir);
3225 		break;
3226 	case PR_CTDIR:
3227 		vap->va_nlink = 2;
3228 		vap->va_size = (avl_numnodes(&p->p_ct_held) + 2) * PRSDSIZE;
3229 		break;
3230 	case PR_TMPLDIR:
3231 		vap->va_nlink = 2;
3232 		vap->va_size = (ct_ntypes + 2) * PRSDSIZE;
3233 		break;
3234 	case PR_AS:
3235 	case PR_PIDFILE:
3236 	case PR_LWPIDFILE:
3237 		if ((p->p_flag & SSYS) || (as = p->p_as) == &kas)
3238 			vap->va_size = 0;
3239 		else
3240 			vap->va_size = as->a_resvsize;
3241 		break;
3242 	case PR_STATUS:
3243 		vap->va_size = PR_OBJSIZE(pstatus32_t, pstatus_t);
3244 		break;
3245 	case PR_LSTATUS:
3246 		vap->va_size = PR_OBJSIZE(prheader32_t, prheader_t) +
3247 		    p->p_lwpcnt * PR_OBJSPAN(lwpstatus32_t, lwpstatus_t);
3248 		break;
3249 	case PR_PSINFO:
3250 		vap->va_size = PR_OBJSIZE(psinfo32_t, psinfo_t);
3251 		break;
3252 	case PR_LPSINFO:
3253 		vap->va_size = PR_OBJSIZE(prheader32_t, prheader_t) +
3254 		    (p->p_lwpcnt + p->p_zombcnt) *
3255 		    PR_OBJSPAN(lwpsinfo32_t, lwpsinfo_t);
3256 		break;
3257 	case PR_MAP:
3258 	case PR_RMAP:
3259 	case PR_XMAP:
3260 		if ((p->p_flag & SSYS) || (as = p->p_as) == &kas)
3261 			vap->va_size = 0;
3262 		else {
3263 			mutex_exit(&p->p_lock);
3264 			AS_LOCK_ENTER(as, RW_WRITER);
3265 			if (type == PR_MAP)
3266 				vap->va_mtime = as->a_updatetime;
3267 			if (type == PR_XMAP)
3268 				vap->va_size = prnsegs(as, 0) *
3269 				    PR_OBJSIZE(prxmap32_t, prxmap_t);
3270 			else
3271 				vap->va_size = prnsegs(as, type == PR_RMAP) *
3272 				    PR_OBJSIZE(prmap32_t, prmap_t);
3273 			AS_LOCK_EXIT(as);
3274 			mutex_enter(&p->p_lock);
3275 		}
3276 		break;
3277 	case PR_CRED:
3278 		mutex_enter(&p->p_crlock);
3279 		vap->va_size = sizeof (prcred_t);
3280 		ngroups = crgetngroups(p->p_cred);
3281 		if (ngroups > 1)
3282 			vap->va_size += (ngroups - 1) * sizeof (gid_t);
3283 		mutex_exit(&p->p_crlock);
3284 		break;
3285 	case PR_PRIV:
3286 		vap->va_size = prgetprivsize();
3287 		break;
3288 	case PR_SECFLAGS:
3289 		vap->va_size = sizeof (prsecflags_t);
3290 		break;
3291 	case PR_SIGACT:
3292 		nsig = PROC_IS_BRANDED(curproc)? BROP(curproc)->b_nsig : NSIG;
3293 		vap->va_size = (nsig-1) *
3294 		    PR_OBJSIZE(struct sigaction32, struct sigaction);
3295 		break;
3296 	case PR_AUXV:
3297 		vap->va_size = __KERN_NAUXV_IMPL * PR_OBJSIZE(auxv32_t, auxv_t);
3298 		break;
3299 #if defined(__x86)
3300 	case PR_LDT:
3301 		mutex_exit(&p->p_lock);
3302 		mutex_enter(&p->p_ldtlock);
3303 		vap->va_size = prnldt(p) * sizeof (struct ssd);
3304 		mutex_exit(&p->p_ldtlock);
3305 		mutex_enter(&p->p_lock);
3306 		break;
3307 #endif
3308 	case PR_USAGE:
3309 		vap->va_size = PR_OBJSIZE(prusage32_t, prusage_t);
3310 		break;
3311 	case PR_LUSAGE:
3312 		vap->va_size = PR_OBJSIZE(prheader32_t, prheader_t) +
3313 		    (p->p_lwpcnt + 1) * PR_OBJSPAN(prusage32_t, prusage_t);
3314 		break;
3315 	case PR_PAGEDATA:
3316 		if ((p->p_flag & SSYS) || (as = p->p_as) == &kas)
3317 			vap->va_size = 0;
3318 		else {
3319 			/*
3320 			 * We can drop p->p_lock before grabbing the
3321 			 * address space lock because p->p_as will not
3322 			 * change while the process is marked P_PR_LOCK.
3323 			 */
3324 			mutex_exit(&p->p_lock);
3325 			AS_LOCK_ENTER(as, RW_WRITER);
3326 #ifdef _LP64
3327 			vap->va_size = iam32bit?
3328 			    prpdsize32(as) : prpdsize(as);
3329 #else
3330 			vap->va_size = prpdsize(as);
3331 #endif
3332 			AS_LOCK_EXIT(as);
3333 			mutex_enter(&p->p_lock);
3334 		}
3335 		break;
3336 	case PR_OPAGEDATA:
3337 		if ((p->p_flag & SSYS) || (as = p->p_as) == &kas)
3338 			vap->va_size = 0;
3339 		else {
3340 			mutex_exit(&p->p_lock);
3341 			AS_LOCK_ENTER(as, RW_WRITER);
3342 #ifdef _LP64
3343 			vap->va_size = iam32bit?
3344 			    oprpdsize32(as) : oprpdsize(as);
3345 #else
3346 			vap->va_size = oprpdsize(as);
3347 #endif
3348 			AS_LOCK_EXIT(as);
3349 			mutex_enter(&p->p_lock);
3350 		}
3351 		break;
3352 	case PR_WATCH:
3353 		vap->va_size = avl_numnodes(&p->p_warea) *
3354 		    PR_OBJSIZE(prwatch32_t, prwatch_t);
3355 		break;
3356 	case PR_LWPSTATUS:
3357 		vap->va_size = PR_OBJSIZE(lwpstatus32_t, lwpstatus_t);
3358 		break;
3359 	case PR_LWPSINFO:
3360 		vap->va_size = PR_OBJSIZE(lwpsinfo32_t, lwpsinfo_t);
3361 		break;
3362 	case PR_LWPUSAGE:
3363 		vap->va_size = PR_OBJSIZE(prusage32_t, prusage_t);
3364 		break;
3365 	case PR_XREGS:
3366 		if (prhasx(p))
3367 			vap->va_size = prgetprxregsize(p);
3368 		else
3369 			vap->va_size = 0;
3370 		break;
3371 	case PR_SPYMASTER:
3372 		if (pnp->pr_common->prc_thread != NULL &&
3373 		    pnp->pr_common->prc_thread->t_lwp->lwp_spymaster != NULL) {
3374 			vap->va_size = PR_OBJSIZE(psinfo32_t, psinfo_t);
3375 		} else {
3376 			vap->va_size = 0;
3377 		}
3378 		break;
3379 #if defined(__sparc)
3380 	case PR_GWINDOWS:
3381 	{
3382 		kthread_t *t;
3383 		int n;
3384 
3385 		/*
3386 		 * If there is no lwp then just make the size zero.
3387 		 * This can happen if the lwp exits between the VOP_LOOKUP()
3388 		 * of the /proc/<pid>/lwp/<lwpid>/gwindows file and the
3389 		 * VOP_GETATTR() of the resulting vnode.
3390 		 */
3391 		if ((t = pcp->prc_thread) == NULL) {
3392 			vap->va_size = 0;
3393 			break;
3394 		}
3395 		/*
3396 		 * Drop p->p_lock while touching the stack.
3397 		 * The P_PR_LOCK flag prevents the lwp from
3398 		 * disappearing while we do this.
3399 		 */
3400 		mutex_exit(&p->p_lock);
3401 		if ((n = prnwindows(ttolwp(t))) == 0)
3402 			vap->va_size = 0;
3403 		else
3404 			vap->va_size = PR_OBJSIZE(gwindows32_t, gwindows_t) -
3405 			    (SPARC_MAXREGWINDOW - n) *
3406 			    PR_OBJSIZE(struct rwindow32, struct rwindow);
3407 		mutex_enter(&p->p_lock);
3408 		break;
3409 	}
3410 	case PR_ASRS:
3411 #ifdef _LP64
3412 		if (p->p_model == DATAMODEL_LP64)
3413 			vap->va_size = sizeof (asrset_t);
3414 		else
3415 #endif
3416 			vap->va_size = 0;
3417 		break;
3418 #endif
3419 	case PR_CTL:
3420 	case PR_LWPCTL:
3421 	default:
3422 		vap->va_size = 0;
3423 		break;
3424 	}
3425 
3426 	prunlock(pnp);
3427 	vap->va_nblocks = (fsblkcnt64_t)btod(vap->va_size);
3428 	return (0);
3429 }
3430 
3431 static int
3432 praccess(vnode_t *vp, int mode, int flags, cred_t *cr, caller_context_t *ct)
3433 {
3434 	prnode_t *pnp = VTOP(vp);
3435 	prnodetype_t type = pnp->pr_type;
3436 	int vmode;
3437 	vtype_t vtype;
3438 	proc_t *p;
3439 	int error = 0;
3440 	vnode_t *rvp;
3441 	vnode_t *xvp;
3442 
3443 	if ((mode & VWRITE) && vn_is_readonly(vp))
3444 		return (EROFS);
3445 
3446 	switch (type) {
3447 	case PR_PROCDIR:
3448 		break;
3449 
3450 	case PR_OBJECT:
3451 	case PR_FD:
3452 		/*
3453 		 * Disallow write access to the underlying objects.
3454 		 * Disallow access to underlying non-regular-file fds.
3455 		 * Disallow access to fds with other than existing open modes.
3456 		 */
3457 		rvp = pnp->pr_realvp;
3458 		vtype = rvp->v_type;
3459 		vmode = pnp->pr_mode;
3460 		if ((type == PR_OBJECT && (mode & VWRITE)) ||
3461 		    (type == PR_FD && vtype != VREG && vtype != VDIR) ||
3462 		    (type == PR_FD && (vmode & mode) != mode &&
3463 		    secpolicy_proc_access(cr) != 0))
3464 			return (EACCES);
3465 		return (VOP_ACCESS(rvp, mode, flags, cr, ct));
3466 
3467 	case PR_PSINFO:		/* these files can be read by anyone */
3468 	case PR_LPSINFO:
3469 	case PR_LWPSINFO:
3470 	case PR_LWPDIR:
3471 	case PR_LWPIDDIR:
3472 	case PR_USAGE:
3473 	case PR_LUSAGE:
3474 	case PR_LWPUSAGE:
3475 		p = pr_p_lock(pnp);
3476 		mutex_exit(&pr_pidlock);
3477 		if (p == NULL)
3478 			return (ENOENT);
3479 		prunlock(pnp);
3480 		break;
3481 
3482 	default:
3483 		/*
3484 		 * Except for the world-readable files above,
3485 		 * only /proc/pid exists if the process is a zombie.
3486 		 */
3487 		if ((error = prlock(pnp,
3488 		    (type == PR_PIDDIR)? ZYES : ZNO)) != 0)
3489 			return (error);
3490 		p = pnp->pr_common->prc_proc;
3491 		if (p != curproc)
3492 			error = priv_proc_cred_perm(cr, p, NULL, mode);
3493 
3494 		if (error != 0 || p == curproc || (p->p_flag & SSYS) ||
3495 		    p->p_as == &kas || (xvp = p->p_exec) == NULL) {
3496 			prunlock(pnp);
3497 		} else {
3498 			/*
3499 			 * Determine if the process's executable is readable.
3500 			 * We have to drop p->p_lock before the secpolicy
3501 			 * and VOP operation.
3502 			 */
3503 			VN_HOLD(xvp);
3504 			prunlock(pnp);
3505 			if (secpolicy_proc_access(cr) != 0)
3506 				error = VOP_ACCESS(xvp, VREAD, 0, cr, ct);
3507 			VN_RELE(xvp);
3508 		}
3509 		if (error)
3510 			return (error);
3511 		break;
3512 	}
3513 
3514 	if (type == PR_CURDIR || type == PR_ROOTDIR) {
3515 		/*
3516 		 * Final access check on the underlying directory vnode.
3517 		 */
3518 		return (VOP_ACCESS(pnp->pr_realvp, mode, flags, cr, ct));
3519 	}
3520 
3521 	/*
3522 	 * Visceral revulsion:  For compatibility with old /proc,
3523 	 * allow the /proc/<pid> directory to be opened for writing.
3524 	 */
3525 	vmode = pnp->pr_mode;
3526 	if (type == PR_PIDDIR)
3527 		vmode |= VWRITE;
3528 	if ((vmode & mode) != mode)
3529 		error = secpolicy_proc_access(cr);
3530 	return (error);
3531 }
3532 
3533 /*
3534  * Array of lookup functions, indexed by /proc file type.
3535  */
3536 static vnode_t *pr_lookup_notdir(), *pr_lookup_procdir(), *pr_lookup_piddir(),
3537 	*pr_lookup_objectdir(), *pr_lookup_lwpdir(), *pr_lookup_lwpiddir(),
3538 	*pr_lookup_fddir(), *pr_lookup_fdinfodir(), *pr_lookup_pathdir(),
3539 	*pr_lookup_tmpldir(), *pr_lookup_ctdir();
3540 
3541 static vnode_t *(*pr_lookup_function[PR_NFILES])() = {
3542 	pr_lookup_procdir,	/* /proc				*/
3543 	pr_lookup_notdir,	/* /proc/self				*/
3544 	pr_lookup_piddir,	/* /proc/<pid>				*/
3545 	pr_lookup_notdir,	/* /proc/<pid>/as			*/
3546 	pr_lookup_notdir,	/* /proc/<pid>/ctl			*/
3547 	pr_lookup_notdir,	/* /proc/<pid>/status			*/
3548 	pr_lookup_notdir,	/* /proc/<pid>/lstatus			*/
3549 	pr_lookup_notdir,	/* /proc/<pid>/psinfo			*/
3550 	pr_lookup_notdir,	/* /proc/<pid>/lpsinfo			*/
3551 	pr_lookup_notdir,	/* /proc/<pid>/map			*/
3552 	pr_lookup_notdir,	/* /proc/<pid>/rmap			*/
3553 	pr_lookup_notdir,	/* /proc/<pid>/xmap			*/
3554 	pr_lookup_notdir,	/* /proc/<pid>/cred			*/
3555 	pr_lookup_notdir,	/* /proc/<pid>/sigact			*/
3556 	pr_lookup_notdir,	/* /proc/<pid>/auxv			*/
3557 #if defined(__x86)
3558 	pr_lookup_notdir,	/* /proc/<pid>/ldt			*/
3559 #endif
3560 	pr_lookup_notdir,	/* /proc/<pid>/usage			*/
3561 	pr_lookup_notdir,	/* /proc/<pid>/lusage			*/
3562 	pr_lookup_notdir,	/* /proc/<pid>/pagedata			*/
3563 	pr_lookup_notdir,	/* /proc/<pid>/watch			*/
3564 	pr_lookup_notdir,	/* /proc/<pid>/cwd			*/
3565 	pr_lookup_notdir,	/* /proc/<pid>/root			*/
3566 	pr_lookup_fddir,	/* /proc/<pid>/fd			*/
3567 	pr_lookup_notdir,	/* /proc/<pid>/fd/nn			*/
3568 	pr_lookup_fdinfodir,	/* /proc/<pid>/fdinfo			*/
3569 	pr_lookup_notdir,	/* /proc/<pid>/fdinfo/nn		*/
3570 	pr_lookup_objectdir,	/* /proc/<pid>/object			*/
3571 	pr_lookup_notdir,	/* /proc/<pid>/object/xxx		*/
3572 	pr_lookup_lwpdir,	/* /proc/<pid>/lwp			*/
3573 	pr_lookup_lwpiddir,	/* /proc/<pid>/lwp/<lwpid>		*/
3574 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpctl	*/
3575 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpname	*/
3576 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpstatus	*/
3577 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpsinfo	*/
3578 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpusage	*/
3579 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/xregs	*/
3580 	pr_lookup_tmpldir,	/* /proc/<pid>/lwp/<lwpid>/templates	*/
3581 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/templates/<id> */
3582 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/spymaster	*/
3583 #if defined(__sparc)
3584 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/gwindows	*/
3585 	pr_lookup_notdir,	/* /proc/<pid>/lwp/<lwpid>/asrs		*/
3586 #endif
3587 	pr_lookup_notdir,	/* /proc/<pid>/priv			*/
3588 	pr_lookup_pathdir,	/* /proc/<pid>/path			*/
3589 	pr_lookup_notdir,	/* /proc/<pid>/path/xxx			*/
3590 	pr_lookup_ctdir,	/* /proc/<pid>/contracts		*/
3591 	pr_lookup_notdir,	/* /proc/<pid>/contracts/<ctid>		*/
3592 	pr_lookup_notdir,	/* /proc/<pid>/secflags			*/
3593 	pr_lookup_notdir,	/* old process file			*/
3594 	pr_lookup_notdir,	/* old lwp file				*/
3595 	pr_lookup_notdir,	/* old pagedata file			*/
3596 };
3597 
3598 static int
3599 prlookup(vnode_t *dp, char *comp, vnode_t **vpp, pathname_t *pathp,
3600     int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
3601     int *direntflags, pathname_t *realpnp)
3602 {
3603 	prnode_t *pnp = VTOP(dp);
3604 	prnodetype_t type = pnp->pr_type;
3605 	int error;
3606 
3607 	ASSERT(dp->v_type == VDIR);
3608 	ASSERT(type < PR_NFILES);
3609 
3610 	if (type != PR_PROCDIR && strcmp(comp, "..") == 0) {
3611 		VN_HOLD(pnp->pr_parent);
3612 		*vpp = pnp->pr_parent;
3613 		return (0);
3614 	}
3615 
3616 	if (*comp == '\0' ||
3617 	    strcmp(comp, ".") == 0 || strcmp(comp, "..") == 0) {
3618 		VN_HOLD(dp);
3619 		*vpp = dp;
3620 		return (0);
3621 	}
3622 
3623 	switch (type) {
3624 	case PR_CURDIR:
3625 	case PR_ROOTDIR:
3626 		/* restrict lookup permission to owner or root */
3627 		if ((error = praccess(dp, VEXEC, 0, cr, ct)) != 0)
3628 			return (error);
3629 		/* FALLTHROUGH */
3630 	case PR_FD:
3631 		/*
3632 		 * Performing a VOP_LOOKUP on the underlying vnode and emitting
3633 		 * the resulting vnode, without encapsulation, as our own is a
3634 		 * very special case when it comes to the assumptions built
3635 		 * into VFS.
3636 		 *
3637 		 * Since the resulting vnode is highly likely to be at some
3638 		 * abitrary position in another filesystem, we insist that the
3639 		 * VTRAVERSE flag is set on the parent.  This prevents things
3640 		 * such as the v_path freshness logic from mistaking the
3641 		 * resulting vnode as a "real" child of the parent, rather than
3642 		 * a consequence of this "procfs wormhole".
3643 		 *
3644 		 * Failure to establish such protections can lead to
3645 		 * incorrectly calculated v_paths being set on nodes reached
3646 		 * through these lookups.
3647 		 */
3648 		ASSERT((dp->v_flag & VTRAVERSE) != 0);
3649 
3650 		dp = pnp->pr_realvp;
3651 		return (VOP_LOOKUP(dp, comp, vpp, pathp, flags, rdir, cr, ct,
3652 		    direntflags, realpnp));
3653 	default:
3654 		break;
3655 	}
3656 
3657 	if ((type == PR_OBJECTDIR || type == PR_FDDIR ||
3658 	    type == PR_FDINFODIR || type == PR_PATHDIR) &&
3659 	    (error = praccess(dp, VEXEC, 0, cr, ct)) != 0)
3660 		return (error);
3661 
3662 	/* XXX - Do we need to pass ct, direntflags, or realpnp? */
3663 	*vpp = (pr_lookup_function[type](dp, comp));
3664 
3665 	return ((*vpp == NULL) ? ENOENT : 0);
3666 }
3667 
3668 /* ARGSUSED */
3669 static int
3670 prcreate(vnode_t *dp, char *comp, vattr_t *vap, vcexcl_t excl,
3671     int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct,
3672     vsecattr_t *vsecp)
3673 {
3674 	int error;
3675 
3676 	if ((error = prlookup(dp, comp, vpp, NULL, 0, NULL, cr,
3677 	    ct, NULL, NULL)) != 0) {
3678 		if (error == ENOENT) {
3679 			/* One can't O_CREAT nonexistent files in /proc. */
3680 			error = EACCES;
3681 		}
3682 		return (error);
3683 	}
3684 
3685 	if (excl == EXCL) {
3686 		/* Disallow the O_EXCL case */
3687 		error = EEXIST;
3688 	} else if ((error = praccess(*vpp, mode, 0, cr, ct)) == 0) {
3689 		/* Before proceeding, handle O_TRUNC if necessary. */
3690 		if (vap->va_mask & AT_SIZE) {
3691 			vnode_t *vp = *vpp;
3692 
3693 			if (vp->v_type == VDIR) {
3694 				/* Only allow O_TRUNC on files */
3695 				error = EISDIR;
3696 			} else if (vp->v_type != VPROC ||
3697 			    VTOP(vp)->pr_type != PR_FD) {
3698 				/*
3699 				 * Disallow for files outside of the
3700 				 * /proc/<pid>/fd/<n> entries
3701 				 */
3702 				error = EACCES;
3703 			} else {
3704 				uint_t mask;
3705 
3706 				vp = VTOP(vp)->pr_realvp;
3707 				mask = vap->va_mask;
3708 				vap->va_mask = AT_SIZE;
3709 				error = VOP_SETATTR(vp, vap, 0, cr, ct);
3710 				vap->va_mask = mask;
3711 			}
3712 		}
3713 	}
3714 
3715 	if (error) {
3716 		VN_RELE(*vpp);
3717 		*vpp = NULL;
3718 	}
3719 	return (error);
3720 }
3721 
3722 /* ARGSUSED */
3723 static vnode_t *
3724 pr_lookup_notdir(vnode_t *dp, char *comp)
3725 {
3726 	return (NULL);
3727 }
3728 
3729 /*
3730  * Find or construct a process vnode for the given pid.
3731  */
3732 static vnode_t *
3733 pr_lookup_procdir(vnode_t *dp, char *comp)
3734 {
3735 	pid_t pid;
3736 	prnode_t *pnp;
3737 	prcommon_t *pcp;
3738 	vnode_t *vp;
3739 	proc_t *p;
3740 	int c;
3741 
3742 	ASSERT(VTOP(dp)->pr_type == PR_PROCDIR);
3743 
3744 	if (strcmp(comp, "self") == 0) {
3745 		pnp = prgetnode(dp, PR_SELF);
3746 		return (PTOV(pnp));
3747 	} else {
3748 		pid = 0;
3749 		while ((c = *comp++) != '\0') {
3750 			if (c < '0' || c > '9')
3751 				return (NULL);
3752 			pid = 10*pid + c - '0';
3753 			if (pid > maxpid)
3754 				return (NULL);
3755 		}
3756 	}
3757 
3758 	pnp = prgetnode(dp, PR_PIDDIR);
3759 
3760 	mutex_enter(&pidlock);
3761 	if ((p = prfind(pid)) == NULL || p->p_stat == SIDL) {
3762 		mutex_exit(&pidlock);
3763 		prfreenode(pnp);
3764 		return (NULL);
3765 	}
3766 	ASSERT(p->p_stat != 0);
3767 
3768 	/* NOTE: we're holding pidlock across the policy call. */
3769 	if (secpolicy_basic_procinfo(CRED(), p, curproc) != 0) {
3770 		mutex_exit(&pidlock);
3771 		prfreenode(pnp);
3772 		return (NULL);
3773 	}
3774 
3775 	mutex_enter(&p->p_lock);
3776 	mutex_exit(&pidlock);
3777 
3778 	/*
3779 	 * If a process vnode already exists and it is not invalid
3780 	 * and it was created by the current process and it belongs
3781 	 * to the same /proc mount point as our parent vnode, then
3782 	 * just use it and discard the newly-allocated prnode.
3783 	 */
3784 	for (vp = p->p_trace; vp != NULL; vp = VTOP(vp)->pr_next) {
3785 		if (!(VTOP(VTOP(vp)->pr_pidfile)->pr_flags & PR_INVAL) &&
3786 		    VTOP(vp)->pr_owner == curproc &&
3787 		    vp->v_vfsp == dp->v_vfsp) {
3788 			ASSERT(!(VTOP(vp)->pr_flags & PR_INVAL));
3789 			VN_HOLD(vp);
3790 			prfreenode(pnp);
3791 			mutex_exit(&p->p_lock);
3792 			return (vp);
3793 		}
3794 	}
3795 	pnp->pr_owner = curproc;
3796 
3797 	/*
3798 	 * prgetnode() initialized most of the prnode.
3799 	 * Finish the job.
3800 	 */
3801 	pcp = pnp->pr_common;	/* the newly-allocated prcommon struct */
3802 	if ((vp = p->p_trace) != NULL) {
3803 		/* discard the new prcommon and use the existing prcommon */
3804 		prfreecommon(pcp);
3805 		pcp = VTOP(vp)->pr_common;
3806 		mutex_enter(&pcp->prc_mutex);
3807 		ASSERT(pcp->prc_refcnt > 0);
3808 		pcp->prc_refcnt++;
3809 		mutex_exit(&pcp->prc_mutex);
3810 		pnp->pr_common = pcp;
3811 	} else {
3812 		/* initialize the new prcommon struct */
3813 		if ((p->p_flag & SSYS) || p->p_as == &kas)
3814 			pcp->prc_flags |= PRC_SYS;
3815 		if (p->p_stat == SZOMB || (p->p_flag & SEXITING) != 0)
3816 			pcp->prc_flags |= PRC_DESTROY;
3817 		pcp->prc_proc = p;
3818 		pcp->prc_datamodel = p->p_model;
3819 		pcp->prc_pid = p->p_pid;
3820 		pcp->prc_slot = p->p_slot;
3821 	}
3822 	pnp->pr_pcommon = pcp;
3823 	pnp->pr_parent = dp;
3824 	VN_HOLD(dp);
3825 	/*
3826 	 * Link in the old, invalid directory vnode so we
3827 	 * can later determine the last close of the file.
3828 	 */
3829 	pnp->pr_next = p->p_trace;
3830 	p->p_trace = dp = PTOV(pnp);
3831 
3832 	/*
3833 	 * Kludge for old /proc: initialize the PR_PIDFILE as well.
3834 	 */
3835 	vp = pnp->pr_pidfile;
3836 	pnp = VTOP(vp);
3837 	pnp->pr_ino = ptoi(pcp->prc_pid);
3838 	pnp->pr_common = pcp;
3839 	pnp->pr_pcommon = pcp;
3840 	pnp->pr_parent = dp;
3841 	pnp->pr_next = p->p_plist;
3842 	p->p_plist = vp;
3843 
3844 	mutex_exit(&p->p_lock);
3845 	return (dp);
3846 }
3847 
3848 static vnode_t *
3849 pr_lookup_piddir(vnode_t *dp, char *comp)
3850 {
3851 	prnode_t *dpnp = VTOP(dp);
3852 	vnode_t *vp;
3853 	prnode_t *pnp;
3854 	proc_t *p;
3855 	user_t *up;
3856 	prdirent_t *dirp;
3857 	int i;
3858 	enum prnodetype type;
3859 
3860 	ASSERT(dpnp->pr_type == PR_PIDDIR);
3861 
3862 	for (i = 0; i < NPIDDIRFILES; i++) {
3863 		/* Skip "." and ".." */
3864 		dirp = &piddir[i+2];
3865 		if (strcmp(comp, dirp->d_name) == 0)
3866 			break;
3867 	}
3868 
3869 	if (i >= NPIDDIRFILES)
3870 		return (NULL);
3871 
3872 	type = (int)dirp->d_ino;
3873 	pnp = prgetnode(dp, type);
3874 
3875 	p = pr_p_lock(dpnp);
3876 	mutex_exit(&pr_pidlock);
3877 	if (p == NULL) {
3878 		prfreenode(pnp);
3879 		return (NULL);
3880 	}
3881 	if (dpnp->pr_pcommon->prc_flags & PRC_DESTROY) {
3882 		switch (type) {
3883 		case PR_PSINFO:
3884 		case PR_USAGE:
3885 			break;
3886 		default:
3887 			prunlock(dpnp);
3888 			prfreenode(pnp);
3889 			return (NULL);
3890 		}
3891 	}
3892 
3893 	switch (type) {
3894 	case PR_CURDIR:
3895 	case PR_ROOTDIR:
3896 		up = PTOU(p);
3897 		vp = (type == PR_CURDIR)? up->u_cdir :
3898 		    (up->u_rdir? up->u_rdir : rootdir);
3899 
3900 		if (vp == NULL) {
3901 			/* can't happen(?) */
3902 			prunlock(dpnp);
3903 			prfreenode(pnp);
3904 			return (NULL);
3905 		}
3906 		/*
3907 		 * Fill in the prnode so future references will
3908 		 * be able to find the underlying object's vnode.
3909 		 */
3910 		VN_HOLD(vp);
3911 		pnp->pr_realvp = vp;
3912 		PTOV(pnp)->v_flag |= VTRAVERSE;
3913 		break;
3914 	default:
3915 		break;
3916 	}
3917 
3918 	mutex_enter(&dpnp->pr_mutex);
3919 
3920 	if ((vp = dpnp->pr_files[i]) != NULL &&
3921 	    !(VTOP(vp)->pr_flags & PR_INVAL)) {
3922 		VN_HOLD(vp);
3923 		mutex_exit(&dpnp->pr_mutex);
3924 		prunlock(dpnp);
3925 		prfreenode(pnp);
3926 		return (vp);
3927 	}
3928 
3929 	/*
3930 	 * prgetnode() initialized most of the prnode.
3931 	 * Finish the job.
3932 	 */
3933 	pnp->pr_common = dpnp->pr_common;
3934 	pnp->pr_pcommon = dpnp->pr_pcommon;
3935 	pnp->pr_parent = dp;
3936 	VN_HOLD(dp);
3937 	pnp->pr_index = i;
3938 
3939 	dpnp->pr_files[i] = vp = PTOV(pnp);
3940 
3941 	/*
3942 	 * Link new vnode into list of all /proc vnodes for the process.
3943 	 */
3944 	if (vp->v_type == VPROC) {
3945 		pnp->pr_next = p->p_plist;
3946 		p->p_plist = vp;
3947 	}
3948 	mutex_exit(&dpnp->pr_mutex);
3949 	prunlock(dpnp);
3950 	return (vp);
3951 }
3952 
3953 static vnode_t *
3954 pr_lookup_objectdir(vnode_t *dp, char *comp)
3955 {
3956 	prnode_t *dpnp = VTOP(dp);
3957 	prnode_t *pnp;
3958 	proc_t *p;
3959 	struct seg *seg;
3960 	struct as *as;
3961 	vnode_t *vp;
3962 	vattr_t vattr;
3963 
3964 	ASSERT(dpnp->pr_type == PR_OBJECTDIR);
3965 
3966 	pnp = prgetnode(dp, PR_OBJECT);
3967 
3968 	if (prlock(dpnp, ZNO) != 0) {
3969 		prfreenode(pnp);
3970 		return (NULL);
3971 	}
3972 	p = dpnp->pr_common->prc_proc;
3973 	if ((p->p_flag & SSYS) || (as = p->p_as) == &kas) {
3974 		prunlock(dpnp);
3975 		prfreenode(pnp);
3976 		return (NULL);
3977 	}
3978 
3979 	/*
3980 	 * We drop p_lock before grabbing the address space lock
3981 	 * in order to avoid a deadlock with the clock thread.
3982 	 * The process will not disappear and its address space
3983 	 * will not change because it is marked P_PR_LOCK.
3984 	 */
3985 	mutex_exit(&p->p_lock);
3986 	AS_LOCK_ENTER(as, RW_READER);
3987 	if ((seg = AS_SEGFIRST(as)) == NULL) {
3988 		vp = NULL;
3989 		goto out;
3990 	}
3991 	if (strcmp(comp, "a.out") == 0) {
3992 		vp = p->p_exec;
3993 		goto out;
3994 	}
3995 	do {
3996 		/*
3997 		 * Manufacture a filename for the "object" directory.
3998 		 */
3999 		vattr.va_mask = AT_FSID|AT_NODEID;
4000 		if (seg->s_ops == &segvn_ops &&
4001 		    SEGOP_GETVP(seg, seg->s_base, &vp) == 0 &&
4002 		    vp != NULL && vp->v_type == VREG &&
4003 		    VOP_GETATTR(vp, &vattr, 0, CRED(), NULL) == 0) {
4004 			char name[64];
4005 
4006 			if (vp == p->p_exec)	/* "a.out" */
4007 				continue;
4008 			pr_object_name(name, vp, &vattr);
4009 			if (strcmp(name, comp) == 0)
4010 				goto out;
4011 		}
4012 	} while ((seg = AS_SEGNEXT(as, seg)) != NULL);
4013 
4014 	vp = NULL;
4015 out:
4016 	if (vp != NULL) {
4017 		VN_HOLD(vp);
4018 	}
4019 	AS_LOCK_EXIT(as);
4020 	mutex_enter(&p->p_lock);
4021 	prunlock(dpnp);
4022 
4023 	if (vp == NULL)
4024 		prfreenode(pnp);
4025 	else {
4026 		/*
4027 		 * Fill in the prnode so future references will
4028 		 * be able to find the underlying object's vnode.
4029 		 * Don't link this prnode into the list of all
4030 		 * prnodes for the process; this is a one-use node.
4031 		 * Its use is entirely to catch and fail opens for writing.
4032 		 */
4033 		pnp->pr_realvp = vp;
4034 		vp = PTOV(pnp);
4035 	}
4036 
4037 	return (vp);
4038 }
4039 
4040 /*
4041  * Find or construct an lwp vnode for the given lwpid.
4042  */
4043 static vnode_t *
4044 pr_lookup_lwpdir(vnode_t *dp, char *comp)
4045 {
4046 	id_t tid;	/* same type as t->t_tid */
4047 	int want_agent;
4048 	prnode_t *dpnp = VTOP(dp);
4049 	prnode_t *pnp;
4050 	prcommon_t *pcp;
4051 	vnode_t *vp;
4052 	proc_t *p;
4053 	kthread_t *t;
4054 	lwpdir_t *ldp;
4055 	lwpent_t *lep;
4056 	int tslot;
4057 	int c;
4058 
4059 	ASSERT(dpnp->pr_type == PR_LWPDIR);
4060 
4061 	tid = 0;
4062 	if (strcmp(comp, "agent") == 0)
4063 		want_agent = 1;
4064 	else {
4065 		want_agent = 0;
4066 		while ((c = *comp++) != '\0') {
4067 			id_t otid;
4068 
4069 			if (c < '0' || c > '9')
4070 				return (NULL);
4071 			otid = tid;
4072 			tid = 10*tid + c - '0';
4073 			if (tid/10 != otid)	/* integer overflow */
4074 				return (NULL);
4075 		}
4076 	}
4077 
4078 	pnp = prgetnode(dp, PR_LWPIDDIR);
4079 
4080 	p = pr_p_lock(dpnp);
4081 	mutex_exit(&pr_pidlock);
4082 	if (p == NULL) {
4083 		prfreenode(pnp);
4084 		return (NULL);
4085 	}
4086 
4087 	if (want_agent) {
4088 		if ((t = p->p_agenttp) == NULL)
4089 			lep = NULL;
4090 		else {
4091 			tid = t->t_tid;
4092 			tslot = t->t_dslot;
4093 			lep = p->p_lwpdir[tslot].ld_entry;
4094 		}
4095 	} else {
4096 		if ((ldp = lwp_hash_lookup(p, tid)) == NULL)
4097 			lep = NULL;
4098 		else {
4099 			tslot = (int)(ldp - p->p_lwpdir);
4100 			lep = ldp->ld_entry;
4101 		}
4102 	}
4103 
4104 	if (lep == NULL) {
4105 		prunlock(dpnp);
4106 		prfreenode(pnp);
4107 		return (NULL);
4108 	}
4109 
4110 	/*
4111 	 * If an lwp vnode already exists and it is not invalid
4112 	 * and it was created by the current process and it belongs
4113 	 * to the same /proc mount point as our parent vnode, then
4114 	 * just use it and discard the newly-allocated prnode.
4115 	 */
4116 	for (vp = lep->le_trace; vp != NULL; vp = VTOP(vp)->pr_next) {
4117 		if (!(VTOP(vp)->pr_flags & PR_INVAL) &&
4118 		    VTOP(vp)->pr_owner == curproc &&
4119 		    vp->v_vfsp == dp->v_vfsp) {
4120 			VN_HOLD(vp);
4121 			prunlock(dpnp);
4122 			prfreenode(pnp);
4123 			return (vp);
4124 		}
4125 	}
4126 	pnp->pr_owner = curproc;
4127 
4128 	/*
4129 	 * prgetnode() initialized most of the prnode.
4130 	 * Finish the job.
4131 	 */
4132 	pcp = pnp->pr_common;	/* the newly-allocated prcommon struct */
4133 	if ((vp = lep->le_trace) != NULL) {
4134 		/* discard the new prcommon and use the existing prcommon */
4135 		prfreecommon(pcp);
4136 		pcp = VTOP(vp)->pr_common;
4137 		mutex_enter(&pcp->prc_mutex);
4138 		ASSERT(pcp->prc_refcnt > 0);
4139 		pcp->prc_refcnt++;
4140 		mutex_exit(&pcp->prc_mutex);
4141 		pnp->pr_common = pcp;
4142 	} else {
4143 		/* initialize the new prcommon struct */
4144 		pcp->prc_flags |= PRC_LWP;
4145 		if ((p->p_flag & SSYS) || p->p_as == &kas)
4146 			pcp->prc_flags |= PRC_SYS;
4147 		if ((t = lep->le_thread) == NULL)
4148 			pcp->prc_flags |= PRC_DESTROY;
4149 		pcp->prc_proc = p;
4150 		pcp->prc_datamodel = dpnp->pr_pcommon->prc_datamodel;
4151 		pcp->prc_pid = p->p_pid;
4152 		pcp->prc_slot = p->p_slot;
4153 		pcp->prc_thread = t;
4154 		pcp->prc_tid = tid;
4155 		pcp->prc_tslot = tslot;
4156 	}
4157 	pnp->pr_pcommon = dpnp->pr_pcommon;
4158 	pnp->pr_parent = dp;
4159 	VN_HOLD(dp);
4160 	/*
4161 	 * Link in the old, invalid directory vnode so we
4162 	 * can later determine the last close of the file.
4163 	 */
4164 	pnp->pr_next = lep->le_trace;
4165 	lep->le_trace = vp = PTOV(pnp);
4166 	prunlock(dpnp);
4167 	return (vp);
4168 }
4169 
4170 static vnode_t *
4171 pr_lookup_lwpiddir(vnode_t *dp, char *comp)
4172 {
4173 	prnode_t *dpnp = VTOP(dp);
4174 	vnode_t *vp;
4175 	prnode_t *pnp;
4176 	proc_t *p;
4177 	prdirent_t *dirp;
4178 	int i;
4179 	enum prnodetype type;
4180 
4181 	ASSERT(dpnp->pr_type == PR_LWPIDDIR);
4182 
4183 	for (i = 0; i < NLWPIDDIRFILES; i++) {
4184 		/* Skip "." and ".." */
4185 		dirp = &lwpiddir[i+2];
4186 		if (strcmp(comp, dirp->d_name) == 0)
4187 			break;
4188 	}
4189 
4190 	if (i >= NLWPIDDIRFILES)
4191 		return (NULL);
4192 
4193 	type = (int)dirp->d_ino;
4194 	pnp = prgetnode(dp, type);
4195 
4196 	p = pr_p_lock(dpnp);
4197 	mutex_exit(&pr_pidlock);
4198 	if (p == NULL) {
4199 		prfreenode(pnp);
4200 		return (NULL);
4201 	}
4202 	if (dpnp->pr_common->prc_flags & PRC_DESTROY) {
4203 		/*
4204 		 * Only the lwpsinfo file is present for zombie lwps.
4205 		 * Nothing is present if the lwp has been reaped.
4206 		 */
4207 		if (dpnp->pr_common->prc_tslot == -1 ||
4208 		    type != PR_LWPSINFO) {
4209 			prunlock(dpnp);
4210 			prfreenode(pnp);
4211 			return (NULL);
4212 		}
4213 	}
4214 
4215 #if defined(__sparc)
4216 	/* the asrs file exists only for sparc v9 _LP64 processes */
4217 	if (type == PR_ASRS && p->p_model != DATAMODEL_LP64) {
4218 		prunlock(dpnp);
4219 		prfreenode(pnp);
4220 		return (NULL);
4221 	}
4222 #endif
4223 
4224 	mutex_enter(&dpnp->pr_mutex);
4225 
4226 	if ((vp = dpnp->pr_files[i]) != NULL &&
4227 	    !(VTOP(vp)->pr_flags & PR_INVAL)) {
4228 		VN_HOLD(vp);
4229 		mutex_exit(&dpnp->pr_mutex);
4230 		prunlock(dpnp);
4231 		prfreenode(pnp);
4232 		return (vp);
4233 	}
4234 
4235 	/*
4236 	 * prgetnode() initialized most of the prnode.
4237 	 * Finish the job.
4238 	 */
4239 	pnp->pr_common = dpnp->pr_common;
4240 	pnp->pr_pcommon = dpnp->pr_pcommon;
4241 	pnp->pr_parent = dp;
4242 	VN_HOLD(dp);
4243 	pnp->pr_index = i;
4244 
4245 	dpnp->pr_files[i] = vp = PTOV(pnp);
4246 
4247 	/*
4248 	 * Link new vnode into list of all /proc vnodes for the process.
4249 	 */
4250 	if (vp->v_type == VPROC) {
4251 		pnp->pr_next = p->p_plist;
4252 		p->p_plist = vp;
4253 	}
4254 	mutex_exit(&dpnp->pr_mutex);
4255 	prunlock(dpnp);
4256 	return (vp);
4257 }
4258 
4259 /*
4260  * Lookup one of the process's file vnodes.
4261  */
4262 static vnode_t *
4263 pr_lookup_fddir(vnode_t *dp, char *comp)
4264 {
4265 	prnode_t *dpnp = VTOP(dp);
4266 	prnode_t *pnp;
4267 	vnode_t *vp = NULL;
4268 	proc_t *p;
4269 	file_t *fp;
4270 	uint_t fd;
4271 	int c;
4272 
4273 	ASSERT(dpnp->pr_type == PR_FDDIR);
4274 
4275 	fd = 0;
4276 	while ((c = *comp++) != '\0') {
4277 		int ofd;
4278 		if (c < '0' || c > '9')
4279 			return (NULL);
4280 		ofd = fd;
4281 		fd = 10 * fd + c - '0';
4282 		if (fd / 10 != ofd)	/* integer overflow */
4283 			return (NULL);
4284 	}
4285 
4286 	pnp = prgetnode(dp, PR_FD);
4287 
4288 	if (prlock(dpnp, ZNO) != 0) {
4289 		prfreenode(pnp);
4290 		return (NULL);
4291 	}
4292 	p = dpnp->pr_common->prc_proc;
4293 	if ((p->p_flag & SSYS) || p->p_as == &kas) {
4294 		prunlock(dpnp);
4295 		prfreenode(pnp);
4296 		return (NULL);
4297 	}
4298 
4299 	if ((fp = pr_getf(p, fd, NULL)) != NULL) {
4300 		pnp->pr_mode = 07111;
4301 		if (fp->f_flag & FREAD)
4302 			pnp->pr_mode |= 0444;
4303 		if (fp->f_flag & FWRITE)
4304 			pnp->pr_mode |= 0222;
4305 		vp = fp->f_vnode;
4306 		VN_HOLD(vp);
4307 	}
4308 
4309 	prunlock(dpnp);
4310 	if (fp != NULL) {
4311 		pr_releasef(fp);
4312 	}
4313 
4314 	if (vp == NULL) {
4315 		prfreenode(pnp);
4316 		return (NULL);
4317 	}
4318 
4319 	/*
4320 	 * Fill in the prnode so future references will
4321 	 * be able to find the underlying object's vnode.
4322 	 * Don't link this prnode into the list of all
4323 	 * prnodes for the process; this is a one-use node.
4324 	 */
4325 	pnp->pr_realvp = vp;
4326 	pnp->pr_parent = dp;		/* needed for prlookup */
4327 	VN_HOLD(dp);
4328 	vp = PTOV(pnp);
4329 	if (pnp->pr_realvp->v_type == VDIR) {
4330 		vp->v_type = VDIR;
4331 		vp->v_flag |= VTRAVERSE;
4332 	}
4333 
4334 	return (vp);
4335 }
4336 
4337 static vnode_t *
4338 pr_lookup_fdinfodir(vnode_t *dp, char *comp)
4339 {
4340 	prnode_t *dpnp = VTOP(dp);
4341 	prnode_t *pnp;
4342 	vnode_t *vp = NULL;
4343 	proc_t *p;
4344 	uint_t fd;
4345 	int c;
4346 
4347 	ASSERT(dpnp->pr_type == PR_FDINFODIR);
4348 
4349 	fd = 0;
4350 	while ((c = *comp++) != '\0') {
4351 		int ofd;
4352 		if (c < '0' || c > '9')
4353 			return (NULL);
4354 		ofd = fd;
4355 		fd = 10 * fd + c - '0';
4356 		if (fd / 10 != ofd)	/* integer overflow */
4357 			return (NULL);
4358 	}
4359 
4360 	pnp = prgetnode(dp, PR_FDINFO);
4361 
4362 	if (prlock(dpnp, ZNO) != 0) {
4363 		prfreenode(pnp);
4364 		return (NULL);
4365 	}
4366 	p = dpnp->pr_common->prc_proc;
4367 	if ((p->p_flag & SSYS) || p->p_as == &kas) {
4368 		prunlock(dpnp);
4369 		prfreenode(pnp);
4370 		return (NULL);
4371 	}
4372 
4373 	/*
4374 	 * Don't link this prnode into the list of all
4375 	 * prnodes for the process; this is a one-use node.
4376 	 * Unlike the FDDIR case, the underlying vnode is not stored in
4377 	 * pnp->pr_realvp. Instead, the fd number is stored in pnp->pr_index
4378 	 * and used by pr_read_fdinfo() to return information for the right
4379 	 * file descriptor.
4380 	 */
4381 	pnp->pr_common = dpnp->pr_common;
4382 	pnp->pr_pcommon = dpnp->pr_pcommon;
4383 	pnp->pr_parent = dp;
4384 	pnp->pr_index = fd;
4385 	VN_HOLD(dp);
4386 	prunlock(dpnp);
4387 	vp = PTOV(pnp);
4388 
4389 	return (vp);
4390 }
4391 
4392 static vnode_t *
4393 pr_lookup_pathdir(vnode_t *dp, char *comp)
4394 {
4395 	prnode_t *dpnp = VTOP(dp);
4396 	prnode_t *pnp;
4397 	vnode_t *vp = NULL;
4398 	proc_t *p;
4399 	uint_t fd, flags = 0;
4400 	int c;
4401 	uf_entry_t *ufp;
4402 	uf_info_t *fip;
4403 	enum { NAME_FD, NAME_OBJECT, NAME_ROOT, NAME_CWD, NAME_UNKNOWN } type;
4404 	char *tmp;
4405 	int idx;
4406 	struct seg *seg;
4407 	struct as *as = NULL;
4408 	vattr_t vattr;
4409 
4410 	ASSERT(dpnp->pr_type == PR_PATHDIR);
4411 
4412 	/*
4413 	 * First, check if this is a numeric entry, in which case we have a
4414 	 * file descriptor.
4415 	 */
4416 	fd = 0;
4417 	type = NAME_FD;
4418 	tmp = comp;
4419 	while ((c = *tmp++) != '\0') {
4420 		int ofd;
4421 		if (c < '0' || c > '9') {
4422 			type = NAME_UNKNOWN;
4423 			break;
4424 		}
4425 		ofd = fd;
4426 		fd = 10*fd + c - '0';
4427 		if (fd/10 != ofd) {	/* integer overflow */
4428 			type = NAME_UNKNOWN;
4429 			break;
4430 		}
4431 	}
4432 
4433 	/*
4434 	 * Next, see if it is one of the special values {root, cwd}.
4435 	 */
4436 	if (type == NAME_UNKNOWN) {
4437 		if (strcmp(comp, "root") == 0)
4438 			type = NAME_ROOT;
4439 		else if (strcmp(comp, "cwd") == 0)
4440 			type = NAME_CWD;
4441 	}
4442 
4443 	/*
4444 	 * Grab the necessary data from the process
4445 	 */
4446 	if (prlock(dpnp, ZNO) != 0)
4447 		return (NULL);
4448 	p = dpnp->pr_common->prc_proc;
4449 
4450 	fip = P_FINFO(p);
4451 
4452 	switch (type) {
4453 	case NAME_ROOT:
4454 		if ((vp = PTOU(p)->u_rdir) == NULL)
4455 			vp = p->p_zone->zone_rootvp;
4456 		VN_HOLD(vp);
4457 		break;
4458 	case NAME_CWD:
4459 		vp = PTOU(p)->u_cdir;
4460 		VN_HOLD(vp);
4461 		break;
4462 	default:
4463 		if ((p->p_flag & SSYS) || (as = p->p_as) == &kas) {
4464 			prunlock(dpnp);
4465 			return (NULL);
4466 		}
4467 	}
4468 	mutex_exit(&p->p_lock);
4469 
4470 	/*
4471 	 * Determine if this is an object entry
4472 	 */
4473 	if (type == NAME_UNKNOWN) {
4474 		/*
4475 		 * Start with the inode index immediately after the number of
4476 		 * files.
4477 		 */
4478 		mutex_enter(&fip->fi_lock);
4479 		idx = fip->fi_nfiles + 4;
4480 		mutex_exit(&fip->fi_lock);
4481 
4482 		if (strcmp(comp, "a.out") == 0) {
4483 			if (p->p_execdir != NULL) {
4484 				vp = p->p_execdir;
4485 				VN_HOLD(vp);
4486 				type = NAME_OBJECT;
4487 				flags |= PR_AOUT;
4488 			} else {
4489 				vp = p->p_exec;
4490 				VN_HOLD(vp);
4491 				type = NAME_OBJECT;
4492 			}
4493 		} else {
4494 			AS_LOCK_ENTER(as, RW_READER);
4495 			if ((seg = AS_SEGFIRST(as)) != NULL) {
4496 				do {
4497 					/*
4498 					 * Manufacture a filename for the
4499 					 * "object" directory.
4500 					 */
4501 					vattr.va_mask = AT_FSID|AT_NODEID;
4502 					if (seg->s_ops == &segvn_ops &&
4503 					    SEGOP_GETVP(seg, seg->s_base, &vp)
4504 					    == 0 &&
4505 					    vp != NULL && vp->v_type == VREG &&
4506 					    VOP_GETATTR(vp, &vattr, 0, CRED(),
4507 					    NULL) == 0) {
4508 						char name[64];
4509 
4510 						if (vp == p->p_exec)
4511 							continue;
4512 						idx++;
4513 						pr_object_name(name, vp,
4514 						    &vattr);
4515 						if (strcmp(name, comp) == 0)
4516 							break;
4517 					}
4518 				} while ((seg = AS_SEGNEXT(as, seg)) != NULL);
4519 			}
4520 
4521 			if (seg == NULL) {
4522 				vp = NULL;
4523 			} else {
4524 				VN_HOLD(vp);
4525 				type = NAME_OBJECT;
4526 			}
4527 
4528 			AS_LOCK_EXIT(as);
4529 		}
4530 	}
4531 
4532 
4533 	switch (type) {
4534 	case NAME_FD:
4535 		mutex_enter(&fip->fi_lock);
4536 		if (fd < fip->fi_nfiles) {
4537 			UF_ENTER(ufp, fip, fd);
4538 			if (ufp->uf_file != NULL) {
4539 				vp = ufp->uf_file->f_vnode;
4540 				VN_HOLD(vp);
4541 			}
4542 			UF_EXIT(ufp);
4543 		}
4544 		mutex_exit(&fip->fi_lock);
4545 		idx = fd + 4;
4546 		break;
4547 	case NAME_ROOT:
4548 		idx = 2;
4549 		break;
4550 	case NAME_CWD:
4551 		idx = 3;
4552 		break;
4553 	case NAME_OBJECT:
4554 	case NAME_UNKNOWN:
4555 		/* Nothing to do */
4556 		break;
4557 	}
4558 
4559 	mutex_enter(&p->p_lock);
4560 	prunlock(dpnp);
4561 
4562 	if (vp != NULL) {
4563 		pnp = prgetnode(dp, PR_PATH);
4564 
4565 		pnp->pr_flags |= flags;
4566 		pnp->pr_common = dpnp->pr_common;
4567 		pnp->pr_pcommon = dpnp->pr_pcommon;
4568 		pnp->pr_realvp = vp;
4569 		pnp->pr_parent = dp;		/* needed for prlookup */
4570 		pnp->pr_ino = pmkino(idx, dpnp->pr_common->prc_slot, PR_PATH);
4571 		VN_HOLD(dp);
4572 		vp = PTOV(pnp);
4573 		vp->v_type = VLNK;
4574 	}
4575 
4576 	return (vp);
4577 }
4578 
4579 /*
4580  * Look up one of the process's active templates.
4581  */
4582 static vnode_t *
4583 pr_lookup_tmpldir(vnode_t *dp, char *comp)
4584 {
4585 	prnode_t *dpnp = VTOP(dp);
4586 	prnode_t *pnp;
4587 	vnode_t *vp = NULL;
4588 	proc_t *p;
4589 	int i;
4590 
4591 	ASSERT(dpnp->pr_type == PR_TMPLDIR);
4592 
4593 	for (i = 0; i < ct_ntypes; i++)
4594 		if (strcmp(comp, ct_types[i]->ct_type_name) == 0)
4595 			break;
4596 	if (i == ct_ntypes)
4597 		return (NULL);
4598 
4599 	pnp = prgetnode(dp, PR_TMPL);
4600 
4601 	if (prlock(dpnp, ZNO) != 0) {
4602 		prfreenode(pnp);
4603 		return (NULL);
4604 	}
4605 	p = dpnp->pr_common->prc_proc;
4606 	if ((p->p_flag & SSYS) || p->p_as == &kas ||
4607 	    (dpnp->pr_common->prc_flags & (PRC_DESTROY | PRC_LWP)) != PRC_LWP) {
4608 		prunlock(dpnp);
4609 		prfreenode(pnp);
4610 		return (NULL);
4611 	}
4612 	if (ttolwp(dpnp->pr_common->prc_thread)->lwp_ct_active[i] != NULL) {
4613 		pnp->pr_common = dpnp->pr_common;
4614 		pnp->pr_pcommon = dpnp->pr_pcommon;
4615 		pnp->pr_parent = dp;
4616 		pnp->pr_cttype = i;
4617 		VN_HOLD(dp);
4618 		vp = PTOV(pnp);
4619 	} else {
4620 		prfreenode(pnp);
4621 	}
4622 	prunlock(dpnp);
4623 
4624 	return (vp);
4625 }
4626 
4627 /*
4628  * Look up one of the contracts owned by the process.
4629  */
4630 static vnode_t *
4631 pr_lookup_ctdir(vnode_t *dp, char *comp)
4632 {
4633 	prnode_t *dpnp = VTOP(dp);
4634 	prnode_t *pnp;
4635 	vnode_t *vp = NULL;
4636 	proc_t *p;
4637 	id_t id = 0;
4638 	contract_t *ct;
4639 	int c;
4640 
4641 	ASSERT(dpnp->pr_type == PR_CTDIR);
4642 
4643 	while ((c = *comp++) != '\0') {
4644 		id_t oid;
4645 		if (c < '0' || c > '9')
4646 			return (NULL);
4647 		oid = id;
4648 		id = 10 * id + c - '0';
4649 		if (id / 10 != oid)	/* integer overflow */
4650 			return (NULL);
4651 	}
4652 
4653 	/*
4654 	 * Search all contracts; we'll filter below.
4655 	 */
4656 	ct = contract_ptr(id, GLOBAL_ZONEUNIQID);
4657 	if (ct == NULL)
4658 		return (NULL);
4659 
4660 	pnp = prgetnode(dp, PR_CT);
4661 
4662 	if (prlock(dpnp, ZNO) != 0) {
4663 		prfreenode(pnp);
4664 		contract_rele(ct);
4665 		return (NULL);
4666 	}
4667 	p = dpnp->pr_common->prc_proc;
4668 	/*
4669 	 * We only allow lookups of contracts owned by this process, or,
4670 	 * if we are zsched and this is a zone's procfs, contracts on
4671 	 * stuff in the zone which are held by processes or contracts
4672 	 * outside the zone.  (see logic in contract_status_common)
4673 	 */
4674 	if ((ct->ct_owner != p) &&
4675 	    !(p == VTOZONE(dp)->zone_zsched && ct->ct_state < CTS_ORPHAN &&
4676 	    VTOZONE(dp)->zone_uniqid == contract_getzuniqid(ct) &&
4677 	    VTOZONE(dp)->zone_uniqid != GLOBAL_ZONEUNIQID &&
4678 	    ct->ct_czuniqid == GLOBAL_ZONEUNIQID)) {
4679 		prunlock(dpnp);
4680 		prfreenode(pnp);
4681 		contract_rele(ct);
4682 		return (NULL);
4683 	}
4684 	pnp->pr_common = dpnp->pr_common;
4685 	pnp->pr_pcommon = dpnp->pr_pcommon;
4686 	pnp->pr_contract = ct;
4687 	pnp->pr_parent = dp;
4688 	pnp->pr_ino = pmkino(id, pnp->pr_common->prc_slot, PR_CT);
4689 	VN_HOLD(dp);
4690 	prunlock(dpnp);
4691 	vp = PTOV(pnp);
4692 
4693 	return (vp);
4694 }
4695 
4696 /*
4697  * Construct an lwp vnode for the old /proc interface.
4698  * We stand on our head to make the /proc plumbing correct.
4699  */
4700 vnode_t *
4701 prlwpnode(prnode_t *pnp, uint_t tid)
4702 {
4703 	char comp[12];
4704 	vnode_t *dp;
4705 	vnode_t *vp;
4706 	prcommon_t *pcp;
4707 	proc_t *p;
4708 
4709 	/*
4710 	 * Lookup the /proc/<pid>/lwp/<lwpid> directory vnode.
4711 	 */
4712 	if (pnp->pr_type == PR_PIDFILE) {
4713 		dp = pnp->pr_parent;		/* /proc/<pid> */
4714 		VN_HOLD(dp);
4715 		vp = pr_lookup_piddir(dp, "lwp");
4716 		VN_RELE(dp);
4717 		if ((dp = vp) == NULL)		/* /proc/<pid>/lwp */
4718 			return (NULL);
4719 	} else if (pnp->pr_type == PR_LWPIDFILE) {
4720 		dp = pnp->pr_parent;		/* /proc/<pid>/lwp/<lwpid> */
4721 		dp = VTOP(dp)->pr_parent;	/* /proc/<pid>/lwp */
4722 		VN_HOLD(dp);
4723 	} else {
4724 		return (NULL);
4725 	}
4726 
4727 	(void) pr_u32tos(tid, comp, sizeof (comp));
4728 	vp = pr_lookup_lwpdir(dp, comp);
4729 	VN_RELE(dp);
4730 	if ((dp = vp) == NULL)
4731 		return (NULL);
4732 
4733 	pnp = prgetnode(dp, PR_LWPIDFILE);
4734 	vp = PTOV(pnp);
4735 
4736 	/*
4737 	 * prgetnode() initialized most of the prnode.
4738 	 * Finish the job.
4739 	 */
4740 	pcp = VTOP(dp)->pr_common;
4741 	pnp->pr_ino = ptoi(pcp->prc_pid);
4742 	pnp->pr_common = pcp;
4743 	pnp->pr_pcommon = VTOP(dp)->pr_pcommon;
4744 	pnp->pr_parent = dp;
4745 	/*
4746 	 * Link new vnode into list of all /proc vnodes for the process.
4747 	 */
4748 	p = pr_p_lock(pnp);
4749 	mutex_exit(&pr_pidlock);
4750 	if (p == NULL) {
4751 		VN_RELE(dp);
4752 		prfreenode(pnp);
4753 		vp = NULL;
4754 	} else if (pcp->prc_thread == NULL) {
4755 		prunlock(pnp);
4756 		VN_RELE(dp);
4757 		prfreenode(pnp);
4758 		vp = NULL;
4759 	} else {
4760 		pnp->pr_next = p->p_plist;
4761 		p->p_plist = vp;
4762 		prunlock(pnp);
4763 	}
4764 
4765 	return (vp);
4766 }
4767 
4768 #if defined(DEBUG)
4769 
4770 static	uint32_t nprnode;
4771 static	uint32_t nprcommon;
4772 
4773 #define	INCREMENT(x)	atomic_inc_32(&x);
4774 #define	DECREMENT(x)	atomic_dec_32(&x);
4775 
4776 #else
4777 
4778 #define	INCREMENT(x)
4779 #define	DECREMENT(x)
4780 
4781 #endif	/* DEBUG */
4782 
4783 /*
4784  * New /proc vnode required; allocate it and fill in most of the fields.
4785  */
4786 prnode_t *
4787 prgetnode(vnode_t *dp, prnodetype_t type)
4788 {
4789 	prnode_t *pnp;
4790 	prcommon_t *pcp;
4791 	vnode_t *vp;
4792 	ulong_t nfiles;
4793 
4794 	INCREMENT(nprnode);
4795 	pnp = kmem_zalloc(sizeof (prnode_t), KM_SLEEP);
4796 
4797 	mutex_init(&pnp->pr_mutex, NULL, MUTEX_DEFAULT, NULL);
4798 	pnp->pr_type = type;
4799 
4800 	pnp->pr_vnode = vn_alloc(KM_SLEEP);
4801 
4802 	vp = PTOV(pnp);
4803 	vp->v_flag = VNOCACHE|VNOMAP|VNOSWAP|VNOMOUNT;
4804 	vn_setops(vp, prvnodeops);
4805 	vp->v_vfsp = dp->v_vfsp;
4806 	vp->v_type = VPROC;
4807 	vp->v_data = (caddr_t)pnp;
4808 
4809 	switch (type) {
4810 	case PR_PIDDIR:
4811 	case PR_LWPIDDIR:
4812 		/*
4813 		 * We need a prcommon and a files array for each of these.
4814 		 */
4815 		INCREMENT(nprcommon);
4816 
4817 		pcp = kmem_zalloc(sizeof (prcommon_t), KM_SLEEP);
4818 		pcp->prc_refcnt = 1;
4819 		pnp->pr_common = pcp;
4820 		mutex_init(&pcp->prc_mutex, NULL, MUTEX_DEFAULT, NULL);
4821 		cv_init(&pcp->prc_wait, NULL, CV_DEFAULT, NULL);
4822 
4823 		nfiles = (type == PR_PIDDIR)? NPIDDIRFILES : NLWPIDDIRFILES;
4824 		pnp->pr_files =
4825 		    kmem_zalloc(nfiles * sizeof (vnode_t *), KM_SLEEP);
4826 
4827 		vp->v_type = VDIR;
4828 		/*
4829 		 * Mode should be read-search by all, but we cannot so long
4830 		 * as we must support compatibility mode with old /proc.
4831 		 * Make /proc/<pid> be read by owner only, search by all.
4832 		 * Make /proc/<pid>/lwp/<lwpid> read-search by all.  Also,
4833 		 * set VDIROPEN on /proc/<pid> so it can be opened for writing.
4834 		 */
4835 		if (type == PR_PIDDIR) {
4836 			/* kludge for old /proc interface */
4837 			prnode_t *xpnp = prgetnode(dp, PR_PIDFILE);
4838 			pnp->pr_pidfile = PTOV(xpnp);
4839 			pnp->pr_mode = 0511;
4840 			vp->v_flag |= VDIROPEN;
4841 		} else {
4842 			pnp->pr_mode = 0555;
4843 		}
4844 
4845 		break;
4846 
4847 	case PR_CURDIR:
4848 	case PR_ROOTDIR:
4849 	case PR_FDDIR:
4850 	case PR_FDINFODIR:
4851 	case PR_OBJECTDIR:
4852 	case PR_PATHDIR:
4853 	case PR_CTDIR:
4854 	case PR_TMPLDIR:
4855 		vp->v_type = VDIR;
4856 		pnp->pr_mode = 0500;	/* read-search by owner only */
4857 		break;
4858 
4859 	case PR_CT:
4860 		vp->v_type = VLNK;
4861 		pnp->pr_mode = 0500;	/* read-search by owner only */
4862 		break;
4863 
4864 	case PR_PATH:
4865 	case PR_SELF:
4866 		vp->v_type = VLNK;
4867 		pnp->pr_mode = 0777;
4868 		break;
4869 
4870 	case PR_LWPDIR:
4871 		vp->v_type = VDIR;
4872 		pnp->pr_mode = 0555;	/* read-search by all */
4873 		break;
4874 
4875 	case PR_AS:
4876 	case PR_TMPL:
4877 		pnp->pr_mode = 0600;	/* read-write by owner only */
4878 		break;
4879 
4880 	case PR_CTL:
4881 	case PR_LWPCTL:
4882 		pnp->pr_mode = 0200;	/* write-only by owner only */
4883 		break;
4884 
4885 	case PR_PIDFILE:
4886 	case PR_LWPIDFILE:
4887 		pnp->pr_mode = 0600;	/* read-write by owner only */
4888 		break;
4889 
4890 	case PR_LWPNAME:
4891 		pnp->pr_mode = 0644;	/* readable by all + owner can write */
4892 		break;
4893 
4894 	case PR_PSINFO:
4895 	case PR_LPSINFO:
4896 	case PR_LWPSINFO:
4897 	case PR_USAGE:
4898 	case PR_LUSAGE:
4899 	case PR_LWPUSAGE:
4900 		pnp->pr_mode = 0444;	/* read-only by all */
4901 		break;
4902 
4903 	default:
4904 		pnp->pr_mode = 0400;	/* read-only by owner only */
4905 		break;
4906 	}
4907 	vn_exists(vp);
4908 	return (pnp);
4909 }
4910 
4911 /*
4912  * Free the storage obtained from prgetnode().
4913  */
4914 void
4915 prfreenode(prnode_t *pnp)
4916 {
4917 	vnode_t *vp;
4918 	ulong_t nfiles;
4919 
4920 	vn_invalid(PTOV(pnp));
4921 	vn_free(PTOV(pnp));
4922 	mutex_destroy(&pnp->pr_mutex);
4923 
4924 	switch (pnp->pr_type) {
4925 	case PR_PIDDIR:
4926 		/* kludge for old /proc interface */
4927 		if (pnp->pr_pidfile != NULL) {
4928 			prfreenode(VTOP(pnp->pr_pidfile));
4929 			pnp->pr_pidfile = NULL;
4930 		}
4931 		/* FALLTHROUGH */
4932 	case PR_LWPIDDIR:
4933 		/*
4934 		 * We allocated a prcommon and a files array for each of these.
4935 		 */
4936 		prfreecommon(pnp->pr_common);
4937 		nfiles = (pnp->pr_type == PR_PIDDIR)?
4938 		    NPIDDIRFILES : NLWPIDDIRFILES;
4939 		kmem_free(pnp->pr_files, nfiles * sizeof (vnode_t *));
4940 		break;
4941 	default:
4942 		break;
4943 	}
4944 	/*
4945 	 * If there is an underlying vnode, be sure
4946 	 * to release it after freeing the prnode.
4947 	 */
4948 	vp = pnp->pr_realvp;
4949 	kmem_free(pnp, sizeof (*pnp));
4950 	DECREMENT(nprnode);
4951 	if (vp != NULL) {
4952 		VN_RELE(vp);
4953 	}
4954 }
4955 
4956 /*
4957  * Free a prcommon structure, if the reference count reaches zero.
4958  */
4959 static void
4960 prfreecommon(prcommon_t *pcp)
4961 {
4962 	mutex_enter(&pcp->prc_mutex);
4963 	ASSERT(pcp->prc_refcnt > 0);
4964 	if (--pcp->prc_refcnt != 0)
4965 		mutex_exit(&pcp->prc_mutex);
4966 	else {
4967 		mutex_exit(&pcp->prc_mutex);
4968 
4969 		ASSERT(pcp->prc_refcnt == 0);
4970 		ASSERT(pcp->prc_selfopens == 0 && pcp->prc_writers == 0);
4971 
4972 		pollhead_clean(&pcp->prc_pollhead);
4973 		mutex_destroy(&pcp->prc_mutex);
4974 		cv_destroy(&pcp->prc_wait);
4975 		kmem_free(pcp, sizeof (prcommon_t));
4976 		DECREMENT(nprcommon);
4977 	}
4978 }
4979 
4980 /*
4981  * Array of readdir functions, indexed by /proc file type.
4982  */
4983 static int pr_readdir_notdir(), pr_readdir_procdir(), pr_readdir_piddir(),
4984 	pr_readdir_objectdir(), pr_readdir_lwpdir(), pr_readdir_lwpiddir(),
4985 	pr_readdir_fddir(), pr_readdir_fdinfodir(), pr_readdir_pathdir(),
4986 	pr_readdir_tmpldir(), pr_readdir_ctdir();
4987 
4988 static int (*pr_readdir_function[PR_NFILES])() = {
4989 	pr_readdir_procdir,	/* /proc				*/
4990 	pr_readdir_notdir,	/* /proc/self				*/
4991 	pr_readdir_piddir,	/* /proc/<pid>				*/
4992 	pr_readdir_notdir,	/* /proc/<pid>/as			*/
4993 	pr_readdir_notdir,	/* /proc/<pid>/ctl			*/
4994 	pr_readdir_notdir,	/* /proc/<pid>/status			*/
4995 	pr_readdir_notdir,	/* /proc/<pid>/lstatus			*/
4996 	pr_readdir_notdir,	/* /proc/<pid>/psinfo			*/
4997 	pr_readdir_notdir,	/* /proc/<pid>/lpsinfo			*/
4998 	pr_readdir_notdir,	/* /proc/<pid>/map			*/
4999 	pr_readdir_notdir,	/* /proc/<pid>/rmap			*/
5000 	pr_readdir_notdir,	/* /proc/<pid>/xmap			*/
5001 	pr_readdir_notdir,	/* /proc/<pid>/cred			*/
5002 	pr_readdir_notdir,	/* /proc/<pid>/sigact			*/
5003 	pr_readdir_notdir,	/* /proc/<pid>/auxv			*/
5004 #if defined(__x86)
5005 	pr_readdir_notdir,	/* /proc/<pid>/ldt			*/
5006 #endif
5007 	pr_readdir_notdir,	/* /proc/<pid>/usage			*/
5008 	pr_readdir_notdir,	/* /proc/<pid>/lusage			*/
5009 	pr_readdir_notdir,	/* /proc/<pid>/pagedata			*/
5010 	pr_readdir_notdir,	/* /proc/<pid>/watch			*/
5011 	pr_readdir_notdir,	/* /proc/<pid>/cwd			*/
5012 	pr_readdir_notdir,	/* /proc/<pid>/root			*/
5013 	pr_readdir_fddir,	/* /proc/<pid>/fd			*/
5014 	pr_readdir_notdir,	/* /proc/<pid>/fd/nn			*/
5015 	pr_readdir_fdinfodir,	/* /proc/<pid>/fdinfo			*/
5016 	pr_readdir_notdir,	/* /proc/<pid>/fdinfo/nn		*/
5017 	pr_readdir_objectdir,	/* /proc/<pid>/object			*/
5018 	pr_readdir_notdir,	/* /proc/<pid>/object/xxx		*/
5019 	pr_readdir_lwpdir,	/* /proc/<pid>/lwp			*/
5020 	pr_readdir_lwpiddir,	/* /proc/<pid>/lwp/<lwpid>		*/
5021 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpctl	*/
5022 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpname	*/
5023 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpstatus	*/
5024 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpsinfo	*/
5025 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/lwpusage	*/
5026 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/xregs	*/
5027 	pr_readdir_tmpldir,	/* /proc/<pid>/lwp/<lwpid>/templates	*/
5028 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/templates/<id> */
5029 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/spymaster	*/
5030 #if defined(__sparc)
5031 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/gwindows	*/
5032 	pr_readdir_notdir,	/* /proc/<pid>/lwp/<lwpid>/asrs		*/
5033 #endif
5034 	pr_readdir_notdir,	/* /proc/<pid>/priv			*/
5035 	pr_readdir_pathdir,	/* /proc/<pid>/path			*/
5036 	pr_readdir_notdir,	/* /proc/<pid>/path/xxx			*/
5037 	pr_readdir_ctdir,	/* /proc/<pid>/contracts		*/
5038 	pr_readdir_notdir,	/* /proc/<pid>/contracts/<ctid>		*/
5039 	pr_readdir_notdir,	/* /proc/<pid>/secflags			*/
5040 	pr_readdir_notdir,	/* old process file			*/
5041 	pr_readdir_notdir,	/* old lwp file				*/
5042 	pr_readdir_notdir,	/* old pagedata file			*/
5043 };
5044 
5045 /* ARGSUSED */
5046 static int
5047 prreaddir(vnode_t *vp, uio_t *uiop, cred_t *cr, int *eofp,
5048     caller_context_t *ct, int flags)
5049 {
5050 	prnode_t *pnp = VTOP(vp);
5051 
5052 	ASSERT(pnp->pr_type < PR_NFILES);
5053 
5054 	/* XXX - Do we need to pass ct and flags? */
5055 	return (pr_readdir_function[pnp->pr_type](pnp, uiop, eofp));
5056 }
5057 
5058 /* ARGSUSED */
5059 static int
5060 pr_readdir_notdir(prnode_t *pnp, uio_t *uiop, int *eofp)
5061 {
5062 	return (ENOTDIR);
5063 }
5064 
5065 /* ARGSUSED */
5066 static int
5067 pr_readdir_procdir(prnode_t *pnp, uio_t *uiop, int *eofp)
5068 {
5069 	zoneid_t zoneid;
5070 	gfs_readdir_state_t gstate;
5071 	int error, eof = 0;
5072 	offset_t n;
5073 
5074 	ASSERT(pnp->pr_type == PR_PROCDIR);
5075 
5076 	zoneid = VTOZONE(PTOV(pnp))->zone_id;
5077 
5078 	if ((error = gfs_readdir_init(&gstate, PNSIZ, PRSDSIZE, uiop,
5079 	    PRROOTINO, PRROOTINO, 0)) != 0)
5080 		return (error);
5081 
5082 	/*
5083 	 * Loop until user's request is satisfied or until all processes
5084 	 * have been examined.
5085 	 */
5086 	while ((error = gfs_readdir_pred(&gstate, uiop, &n)) == 0) {
5087 		uint_t pid;
5088 		int pslot;
5089 		proc_t *p;
5090 
5091 		/*
5092 		 * Find next entry.  Skip processes not visible where
5093 		 * this /proc was mounted.
5094 		 */
5095 		mutex_enter(&pidlock);
5096 		while (n < v.v_proc &&
5097 		    ((p = pid_entry(n)) == NULL || p->p_stat == SIDL ||
5098 		    (zoneid != GLOBAL_ZONEID && p->p_zone->zone_id != zoneid) ||
5099 		    secpolicy_basic_procinfo(CRED(), p, curproc) != 0))
5100 			n++;
5101 
5102 		/*
5103 		 * Stop when entire proc table has been examined.
5104 		 */
5105 		if (n >= v.v_proc) {
5106 			mutex_exit(&pidlock);
5107 			eof = 1;
5108 			break;
5109 		}
5110 
5111 		ASSERT(p->p_stat != 0);
5112 		pid = p->p_pid;
5113 		pslot = p->p_slot;
5114 		mutex_exit(&pidlock);
5115 		error = gfs_readdir_emitn(&gstate, uiop, n,
5116 		    pmkino(0, pslot, PR_PIDDIR), pid);
5117 		if (error)
5118 			break;
5119 	}
5120 
5121 	return (gfs_readdir_fini(&gstate, error, eofp, eof));
5122 }
5123 
5124 /* ARGSUSED */
5125 static int
5126 pr_readdir_piddir(prnode_t *pnp, uio_t *uiop, int *eofp)
5127 {
5128 	int zombie = ((pnp->pr_pcommon->prc_flags & PRC_DESTROY) != 0);
5129 	prdirent_t dirent;
5130 	prdirent_t *dirp;
5131 	offset_t off;
5132 	int error;
5133 
5134 	ASSERT(pnp->pr_type == PR_PIDDIR);
5135 
5136 	if (uiop->uio_offset < 0 ||
5137 	    uiop->uio_offset % sizeof (prdirent_t) != 0 ||
5138 	    uiop->uio_resid < sizeof (prdirent_t))
5139 		return (EINVAL);
5140 	if (pnp->pr_pcommon->prc_proc == NULL)
5141 		return (ENOENT);
5142 	if (uiop->uio_offset >= sizeof (piddir))
5143 		goto out;
5144 
5145 	/*
5146 	 * Loop until user's request is satisfied, omitting some
5147 	 * files along the way if the process is a zombie.
5148 	 */
5149 	for (dirp = &piddir[uiop->uio_offset / sizeof (prdirent_t)];
5150 	    uiop->uio_resid >= sizeof (prdirent_t) &&
5151 	    dirp < &piddir[NPIDDIRFILES+2];
5152 	    uiop->uio_offset = off + sizeof (prdirent_t), dirp++) {
5153 		off = uiop->uio_offset;
5154 		if (zombie) {
5155 			switch (dirp->d_ino) {
5156 			case PR_PIDDIR:
5157 			case PR_PROCDIR:
5158 			case PR_PSINFO:
5159 			case PR_USAGE:
5160 				break;
5161 			default:
5162 				continue;
5163 			}
5164 		}
5165 		bcopy(dirp, &dirent, sizeof (prdirent_t));
5166 		if (dirent.d_ino == PR_PROCDIR)
5167 			dirent.d_ino = PRROOTINO;
5168 		else
5169 			dirent.d_ino = pmkino(0, pnp->pr_pcommon->prc_slot,
5170 			    dirent.d_ino);
5171 		if ((error = uiomove((caddr_t)&dirent, sizeof (prdirent_t),
5172 		    UIO_READ, uiop)) != 0)
5173 			return (error);
5174 	}
5175 out:
5176 	if (eofp)
5177 		*eofp = (uiop->uio_offset >= sizeof (piddir));
5178 	return (0);
5179 }
5180 
5181 static void
5182 rebuild_objdir(struct as *as)
5183 {
5184 	struct seg *seg;
5185 	vnode_t *vp;
5186 	vattr_t vattr;
5187 	vnode_t **dir;
5188 	ulong_t nalloc;
5189 	ulong_t nentries;
5190 	int i, j;
5191 	ulong_t nold, nnew;
5192 
5193 	ASSERT(AS_WRITE_HELD(as));
5194 
5195 	if (as->a_updatedir == 0 && as->a_objectdir != NULL)
5196 		return;
5197 	as->a_updatedir = 0;
5198 
5199 	if ((nalloc = avl_numnodes(&as->a_segtree)) == 0 ||
5200 	    (seg = AS_SEGFIRST(as)) == NULL)	/* can't happen? */
5201 		return;
5202 
5203 	/*
5204 	 * Allocate space for the new object directory.
5205 	 * (This is usually about two times too many entries.)
5206 	 */
5207 	nalloc = (nalloc + 0xf) & ~0xf;		/* multiple of 16 */
5208 	dir = kmem_zalloc(nalloc * sizeof (vnode_t *), KM_SLEEP);
5209 
5210 	/* fill in the new directory with desired entries */
5211 	nentries = 0;
5212 	do {
5213 		vattr.va_mask = AT_FSID|AT_NODEID;
5214 		if (seg->s_ops == &segvn_ops &&
5215 		    SEGOP_GETVP(seg, seg->s_base, &vp) == 0 &&
5216 		    vp != NULL && vp->v_type == VREG &&
5217 		    VOP_GETATTR(vp, &vattr, 0, CRED(), NULL) == 0) {
5218 			for (i = 0; i < nentries; i++)
5219 				if (vp == dir[i])
5220 					break;
5221 			if (i == nentries) {
5222 				ASSERT(nentries < nalloc);
5223 				dir[nentries++] = vp;
5224 			}
5225 		}
5226 	} while ((seg = AS_SEGNEXT(as, seg)) != NULL);
5227 
5228 	if (as->a_objectdir == NULL) {	/* first time */
5229 		as->a_objectdir = dir;
5230 		as->a_sizedir = nalloc;
5231 		return;
5232 	}
5233 
5234 	/*
5235 	 * Null out all of the defunct entries in the old directory.
5236 	 */
5237 	nold = 0;
5238 	nnew = nentries;
5239 	for (i = 0; i < as->a_sizedir; i++) {
5240 		if ((vp = as->a_objectdir[i]) != NULL) {
5241 			for (j = 0; j < nentries; j++) {
5242 				if (vp == dir[j]) {
5243 					dir[j] = NULL;
5244 					nnew--;
5245 					break;
5246 				}
5247 			}
5248 			if (j == nentries)
5249 				as->a_objectdir[i] = NULL;
5250 			else
5251 				nold++;
5252 		}
5253 	}
5254 
5255 	if (nold + nnew > as->a_sizedir) {
5256 		/*
5257 		 * Reallocate the old directory to have enough
5258 		 * space for the old and new entries combined.
5259 		 * Round up to the next multiple of 16.
5260 		 */
5261 		ulong_t newsize = (nold + nnew + 0xf) & ~0xf;
5262 		vnode_t **newdir = kmem_zalloc(newsize * sizeof (vnode_t *),
5263 		    KM_SLEEP);
5264 		bcopy(as->a_objectdir, newdir,
5265 		    as->a_sizedir * sizeof (vnode_t *));
5266 		kmem_free(as->a_objectdir, as->a_sizedir * sizeof (vnode_t *));
5267 		as->a_objectdir = newdir;
5268 		as->a_sizedir = newsize;
5269 	}
5270 
5271 	/*
5272 	 * Move all new entries to the old directory and
5273 	 * deallocate the space used by the new directory.
5274 	 */
5275 	if (nnew) {
5276 		for (i = 0, j = 0; i < nentries; i++) {
5277 			if ((vp = dir[i]) == NULL)
5278 				continue;
5279 			for (; j < as->a_sizedir; j++) {
5280 				if (as->a_objectdir[j] != NULL)
5281 					continue;
5282 				as->a_objectdir[j++] = vp;
5283 				break;
5284 			}
5285 		}
5286 	}
5287 	kmem_free(dir, nalloc * sizeof (vnode_t *));
5288 }
5289 
5290 /*
5291  * Return the vnode from a slot in the process's object directory.
5292  * The caller must have locked the process's address space.
5293  * The only caller is below, in pr_readdir_objectdir().
5294  */
5295 static vnode_t *
5296 obj_entry(struct as *as, int slot)
5297 {
5298 	ASSERT(AS_LOCK_HELD(as));
5299 	if (as->a_objectdir == NULL)
5300 		return (NULL);
5301 	ASSERT(slot < as->a_sizedir);
5302 	return (as->a_objectdir[slot]);
5303 }
5304 
5305 /* ARGSUSED */
5306 static int
5307 pr_readdir_objectdir(prnode_t *pnp, uio_t *uiop, int *eofp)
5308 {
5309 	gfs_readdir_state_t gstate;
5310 	int error, eof = 0;
5311 	offset_t n;
5312 	int pslot;
5313 	size_t objdirsize;
5314 	proc_t *p;
5315 	struct as *as;
5316 	vnode_t *vp;
5317 
5318 	ASSERT(pnp->pr_type == PR_OBJECTDIR);
5319 
5320 	if ((error = prlock(pnp, ZNO)) != 0)
5321 		return (error);
5322 	p = pnp->pr_common->prc_proc;
5323 	pslot = p->p_slot;
5324 
5325 	/*
5326 	 * We drop p_lock before grabbing the address space lock
5327 	 * in order to avoid a deadlock with the clock thread.
5328 	 * The process will not disappear and its address space
5329 	 * will not change because it is marked P_PR_LOCK.
5330 	 */
5331 	mutex_exit(&p->p_lock);
5332 
5333 	if ((error = gfs_readdir_init(&gstate, 64, PRSDSIZE, uiop,
5334 	    pmkino(0, pslot, PR_PIDDIR),
5335 	    pmkino(0, pslot, PR_OBJECTDIR), 0)) != 0) {
5336 		mutex_enter(&p->p_lock);
5337 		prunlock(pnp);
5338 		return (error);
5339 	}
5340 
5341 	if ((p->p_flag & SSYS) || (as = p->p_as) == &kas) {
5342 		as = NULL;
5343 		objdirsize = 0;
5344 	}
5345 
5346 	/*
5347 	 * Loop until user's request is satisfied or until
5348 	 * all mapped objects have been examined. Cannot hold
5349 	 * the address space lock for the following call as
5350 	 * gfs_readdir_pred() utimately causes a call to uiomove().
5351 	 */
5352 	while ((error = gfs_readdir_pred(&gstate, uiop, &n)) == 0) {
5353 		vattr_t vattr;
5354 		char str[64];
5355 
5356 		/*
5357 		 * Set the correct size of the directory just
5358 		 * in case the process has changed it's address
5359 		 * space via mmap/munmap calls.
5360 		 */
5361 		if (as != NULL) {
5362 			AS_LOCK_ENTER(as, RW_WRITER);
5363 			if (as->a_updatedir)
5364 				rebuild_objdir(as);
5365 			objdirsize = as->a_sizedir;
5366 		}
5367 
5368 		/*
5369 		 * Find next object.
5370 		 */
5371 		vattr.va_mask = AT_FSID | AT_NODEID;
5372 		while (n < objdirsize && (((vp = obj_entry(as, n)) == NULL) ||
5373 		    (VOP_GETATTR(vp, &vattr, 0, CRED(), NULL)
5374 		    != 0))) {
5375 			vattr.va_mask = AT_FSID | AT_NODEID;
5376 			n++;
5377 		}
5378 
5379 		if (as != NULL)
5380 			AS_LOCK_EXIT(as);
5381 
5382 		/*
5383 		 * Stop when all objects have been reported.
5384 		 */
5385 		if (n >= objdirsize) {
5386 			eof = 1;
5387 			break;
5388 		}
5389 
5390 		if (vp == p->p_exec)
5391 			(void) strcpy(str, "a.out");
5392 		else
5393 			pr_object_name(str, vp, &vattr);
5394 
5395 		error = gfs_readdir_emit(&gstate, uiop, n, vattr.va_nodeid,
5396 		    str, 0);
5397 
5398 		if (error)
5399 			break;
5400 	}
5401 
5402 	mutex_enter(&p->p_lock);
5403 	prunlock(pnp);
5404 
5405 	return (gfs_readdir_fini(&gstate, error, eofp, eof));
5406 }
5407 
5408 /* ARGSUSED */
5409 static int
5410 pr_readdir_lwpdir(prnode_t *pnp, uio_t *uiop, int *eofp)
5411 {
5412 	gfs_readdir_state_t gstate;
5413 	int error, eof = 0;
5414 	offset_t tslot;
5415 	proc_t *p;
5416 	int pslot;
5417 	lwpdir_t *lwpdir;
5418 	int lwpdirsize;
5419 
5420 	ASSERT(pnp->pr_type == PR_LWPDIR);
5421 
5422 	p = pr_p_lock(pnp);
5423 	mutex_exit(&pr_pidlock);
5424 	if (p == NULL)
5425 		return (ENOENT);
5426 	ASSERT(p == pnp->pr_common->prc_proc);
5427 	pslot = p->p_slot;
5428 	lwpdir = p->p_lwpdir;
5429 	lwpdirsize = p->p_lwpdir_sz;
5430 
5431 	/*
5432 	 * Drop p->p_lock so we can safely do uiomove().
5433 	 * The lwp directory will not change because
5434 	 * we have the process locked with P_PR_LOCK.
5435 	 */
5436 	mutex_exit(&p->p_lock);
5437 
5438 
5439 	if ((error = gfs_readdir_init(&gstate, PLNSIZ, PRSDSIZE, uiop,
5440 	    pmkino(0, pslot, PR_PIDDIR),
5441 	    pmkino(0, pslot, PR_LWPDIR), 0)) != 0) {
5442 		mutex_enter(&p->p_lock);
5443 		prunlock(pnp);
5444 		return (error);
5445 	}
5446 
5447 	/*
5448 	 * Loop until user's request is satisfied or until all lwps
5449 	 * have been examined.
5450 	 */
5451 	while ((error = gfs_readdir_pred(&gstate, uiop, &tslot)) == 0) {
5452 		lwpent_t *lep;
5453 		uint_t tid;
5454 
5455 		/*
5456 		 * Find next LWP.
5457 		 */
5458 		while (tslot < lwpdirsize &&
5459 		    ((lep = lwpdir[tslot].ld_entry) == NULL))
5460 			tslot++;
5461 		/*
5462 		 * Stop when all lwps have been reported.
5463 		 */
5464 		if (tslot >= lwpdirsize) {
5465 			eof = 1;
5466 			break;
5467 		}
5468 
5469 		tid = lep->le_lwpid;
5470 		error = gfs_readdir_emitn(&gstate, uiop, tslot,
5471 		    pmkino(tslot, pslot, PR_LWPIDDIR), tid);
5472 		if (error)
5473 			break;
5474 	}
5475 
5476 	mutex_enter(&p->p_lock);
5477 	prunlock(pnp);
5478 
5479 	return (gfs_readdir_fini(&gstate, error, eofp, eof));
5480 }
5481 
5482 /* ARGSUSED */
5483 static int
5484 pr_readdir_lwpiddir(prnode_t *pnp, uio_t *uiop, int *eofp)
5485 {
5486 	prcommon_t *pcp = pnp->pr_common;
5487 	int zombie = ((pcp->prc_flags & PRC_DESTROY) != 0);
5488 	prdirent_t dirent;
5489 	prdirent_t *dirp;
5490 	offset_t off;
5491 	int error;
5492 	int pslot;
5493 	int tslot;
5494 
5495 	ASSERT(pnp->pr_type == PR_LWPIDDIR);
5496 
5497 	if (uiop->uio_offset < 0 ||
5498 	    uiop->uio_offset % sizeof (prdirent_t) != 0 ||
5499 	    uiop->uio_resid < sizeof (prdirent_t))
5500 		return (EINVAL);
5501 	if (pcp->prc_proc == NULL || pcp->prc_tslot == -1)
5502 		return (ENOENT);
5503 	if (uiop->uio_offset >= sizeof (lwpiddir))
5504 		goto out;
5505 
5506 	/*
5507 	 * Loop until user's request is satisfied, omitting some files
5508 	 * along the way if the lwp is a zombie and also depending
5509 	 * on the data model of the process.
5510 	 */
5511 	pslot = pcp->prc_slot;
5512 	tslot = pcp->prc_tslot;
5513 	for (dirp = &lwpiddir[uiop->uio_offset / sizeof (prdirent_t)];
5514 	    uiop->uio_resid >= sizeof (prdirent_t) &&
5515 	    dirp < &lwpiddir[NLWPIDDIRFILES+2];
5516 	    uiop->uio_offset = off + sizeof (prdirent_t), dirp++) {
5517 		off = uiop->uio_offset;
5518 		if (zombie) {
5519 			switch (dirp->d_ino) {
5520 			case PR_LWPIDDIR:
5521 			case PR_LWPDIR:
5522 			case PR_LWPSINFO:
5523 				break;
5524 			default:
5525 				continue;
5526 			}
5527 		}
5528 #if defined(__sparc)
5529 		/* the asrs file exists only for sparc v9 _LP64 processes */
5530 		if (dirp->d_ino == PR_ASRS &&
5531 		    pcp->prc_datamodel != DATAMODEL_LP64)
5532 			continue;
5533 #endif
5534 		bcopy(dirp, &dirent, sizeof (prdirent_t));
5535 		if (dirent.d_ino == PR_LWPDIR)
5536 			dirent.d_ino = pmkino(0, pslot, dirp->d_ino);
5537 		else
5538 			dirent.d_ino = pmkino(tslot, pslot, dirp->d_ino);
5539 		if ((error = uiomove((caddr_t)&dirent, sizeof (prdirent_t),
5540 		    UIO_READ, uiop)) != 0)
5541 			return (error);
5542 	}
5543 out:
5544 	if (eofp)
5545 		*eofp = (uiop->uio_offset >= sizeof (lwpiddir));
5546 	return (0);
5547 }
5548 
5549 /*
5550  * Helper function for reading a directory which lists open file desciptors
5551  */
5552 static int
5553 pr_readdir_fdlist(prnode_t *pnp, uio_t *uiop, int *eofp,
5554     prnodetype_t dirtype, prnodetype_t entrytype)
5555 {
5556 	gfs_readdir_state_t gstate;
5557 	int error, eof = 0;
5558 	offset_t n;
5559 	proc_t *p;
5560 	int pslot;
5561 	int fddirsize;
5562 	uf_info_t *fip;
5563 
5564 	if ((error = prlock(pnp, ZNO)) != 0)
5565 		return (error);
5566 	p = pnp->pr_common->prc_proc;
5567 	pslot = p->p_slot;
5568 	fip = P_FINFO(p);
5569 	mutex_exit(&p->p_lock);
5570 
5571 	if ((error = gfs_readdir_init(&gstate, PLNSIZ, PRSDSIZE, uiop,
5572 	    pmkino(0, pslot, PR_PIDDIR), pmkino(0, pslot, dirtype), 0)) != 0) {
5573 		mutex_enter(&p->p_lock);
5574 		prunlock(pnp);
5575 		return (error);
5576 	}
5577 
5578 	mutex_enter(&fip->fi_lock);
5579 	if ((p->p_flag & SSYS) || p->p_as == &kas)
5580 		fddirsize = 0;
5581 	else
5582 		fddirsize = fip->fi_nfiles;
5583 
5584 	/*
5585 	 * Loop until user's request is satisfied or until
5586 	 * all file descriptors have been examined.
5587 	 */
5588 	while ((error = gfs_readdir_pred(&gstate, uiop, &n)) == 0) {
5589 		/*
5590 		 * Find next fd.
5591 		 */
5592 		while (n < fddirsize && fip->fi_list[n].uf_file == NULL)
5593 			n++;
5594 		/*
5595 		 * Stop when all fds have been reported.
5596 		 */
5597 		if (n >= fddirsize) {
5598 			eof = 1;
5599 			break;
5600 		}
5601 
5602 		error = gfs_readdir_emitn(&gstate, uiop, n,
5603 		    pmkino(n, pslot, entrytype), n);
5604 		if (error)
5605 			break;
5606 	}
5607 
5608 	mutex_exit(&fip->fi_lock);
5609 	mutex_enter(&p->p_lock);
5610 	prunlock(pnp);
5611 
5612 	return (gfs_readdir_fini(&gstate, error, eofp, eof));
5613 }
5614 
5615 static int
5616 pr_readdir_fddir(prnode_t *pnp, uio_t *uiop, int *eofp)
5617 {
5618 
5619 	ASSERT(pnp->pr_type == PR_FDDIR);
5620 
5621 	return (pr_readdir_fdlist(pnp, uiop, eofp, pnp->pr_type, PR_FD));
5622 }
5623 
5624 static int
5625 pr_readdir_fdinfodir(prnode_t *pnp, uio_t *uiop, int *eofp)
5626 {
5627 
5628 	ASSERT(pnp->pr_type == PR_FDINFODIR);
5629 
5630 	return (pr_readdir_fdlist(pnp, uiop, eofp, pnp->pr_type, PR_FDINFO));
5631 }
5632 
5633 /* ARGSUSED */
5634 static int
5635 pr_readdir_pathdir(prnode_t *pnp, uio_t *uiop, int *eofp)
5636 {
5637 	longlong_t bp[DIRENT64_RECLEN(64) / sizeof (longlong_t)];
5638 	dirent64_t *dirent = (dirent64_t *)bp;
5639 	int reclen;
5640 	ssize_t oresid;
5641 	offset_t off, idx;
5642 	int error = 0;
5643 	proc_t *p;
5644 	int fd, obj;
5645 	int pslot;
5646 	int fddirsize;
5647 	uf_info_t *fip;
5648 	struct as *as = NULL;
5649 	size_t objdirsize;
5650 	vattr_t vattr;
5651 	vnode_t *vp;
5652 
5653 	ASSERT(pnp->pr_type == PR_PATHDIR);
5654 
5655 	if (uiop->uio_offset < 0 ||
5656 	    uiop->uio_resid <= 0 ||
5657 	    (uiop->uio_offset % PRSDSIZE) != 0)
5658 		return (EINVAL);
5659 	oresid = uiop->uio_resid;
5660 	bzero(bp, sizeof (bp));
5661 
5662 	if ((error = prlock(pnp, ZNO)) != 0)
5663 		return (error);
5664 	p = pnp->pr_common->prc_proc;
5665 	fip = P_FINFO(p);
5666 	pslot = p->p_slot;
5667 	mutex_exit(&p->p_lock);
5668 
5669 	if ((p->p_flag & SSYS) || (as = p->p_as) == &kas) {
5670 		as = NULL;
5671 		objdirsize = 0;
5672 	} else {
5673 		AS_LOCK_ENTER(as, RW_WRITER);
5674 		if (as->a_updatedir)
5675 			rebuild_objdir(as);
5676 		objdirsize = as->a_sizedir;
5677 		AS_LOCK_EXIT(as);
5678 		as = NULL;
5679 	}
5680 
5681 	mutex_enter(&fip->fi_lock);
5682 	if ((p->p_flag & SSYS) || p->p_as == &kas)
5683 		fddirsize = 0;
5684 	else
5685 		fddirsize = fip->fi_nfiles;
5686 
5687 	for (; uiop->uio_resid > 0; uiop->uio_offset = off + PRSDSIZE) {
5688 		/*
5689 		 * There are 4 special files in the path directory: ".", "..",
5690 		 * "root", and "cwd".  We handle those specially here.
5691 		 */
5692 		off = uiop->uio_offset;
5693 		idx = off / PRSDSIZE;
5694 		if (off == 0) {				/* "." */
5695 			dirent->d_ino = pmkino(0, pslot, PR_PATHDIR);
5696 			dirent->d_name[0] = '.';
5697 			dirent->d_name[1] = '\0';
5698 			reclen = DIRENT64_RECLEN(1);
5699 		} else if (idx == 1) {			/* ".." */
5700 			dirent->d_ino = pmkino(0, pslot, PR_PIDDIR);
5701 			dirent->d_name[0] = '.';
5702 			dirent->d_name[1] = '.';
5703 			dirent->d_name[2] = '\0';
5704 			reclen = DIRENT64_RECLEN(2);
5705 		} else if (idx == 2) {			/* "root" */
5706 			dirent->d_ino = pmkino(idx, pslot, PR_PATH);
5707 			(void) strcpy(dirent->d_name, "root");
5708 			reclen = DIRENT64_RECLEN(4);
5709 		} else if (idx == 3) {			/* "cwd" */
5710 			dirent->d_ino = pmkino(idx, pslot, PR_PATH);
5711 			(void) strcpy(dirent->d_name, "cwd");
5712 			reclen = DIRENT64_RECLEN(3);
5713 		} else if (idx < 4 + fddirsize) {
5714 			/*
5715 			 * In this case, we have one of the file descriptors.
5716 			 */
5717 			fd = idx - 4;
5718 			if (fip->fi_list[fd].uf_file == NULL)
5719 				continue;
5720 			dirent->d_ino = pmkino(idx, pslot, PR_PATH);
5721 			(void) pr_u32tos(fd, dirent->d_name, PLNSIZ+1);
5722 			reclen = DIRENT64_RECLEN(PLNSIZ);
5723 		} else if (idx < 4 + fddirsize + objdirsize) {
5724 			if (fip != NULL) {
5725 				mutex_exit(&fip->fi_lock);
5726 				fip = NULL;
5727 			}
5728 
5729 			/*
5730 			 * We drop p_lock before grabbing the address space lock
5731 			 * in order to avoid a deadlock with the clock thread.
5732 			 * The process will not disappear and its address space
5733 			 * will not change because it is marked P_PR_LOCK.
5734 			 */
5735 			if (as == NULL) {
5736 				as = p->p_as;
5737 				AS_LOCK_ENTER(as, RW_WRITER);
5738 			}
5739 
5740 			if (as->a_updatedir) {
5741 				rebuild_objdir(as);
5742 				objdirsize = as->a_sizedir;
5743 			}
5744 
5745 			obj = idx - 4 - fddirsize;
5746 			if ((vp = obj_entry(as, obj)) == NULL)
5747 				continue;
5748 			vattr.va_mask = AT_FSID|AT_NODEID;
5749 			if (VOP_GETATTR(vp, &vattr, 0, CRED(), NULL) != 0)
5750 				continue;
5751 			if (vp == p->p_exec)
5752 				(void) strcpy(dirent->d_name, "a.out");
5753 			else
5754 				pr_object_name(dirent->d_name, vp, &vattr);
5755 			dirent->d_ino = pmkino(idx, pslot, PR_PATH);
5756 			reclen = DIRENT64_RECLEN(strlen(dirent->d_name));
5757 		} else {
5758 			break;
5759 		}
5760 
5761 		dirent->d_off = uiop->uio_offset + PRSDSIZE;
5762 		dirent->d_reclen = (ushort_t)reclen;
5763 		if (reclen > uiop->uio_resid) {
5764 			/*
5765 			 * Error if no entries have been returned yet.
5766 			 */
5767 			if (uiop->uio_resid == oresid)
5768 				error = EINVAL;
5769 			break;
5770 		}
5771 		/*
5772 		 * Drop the address space lock to do the uiomove().
5773 		 */
5774 		if (as != NULL)
5775 			AS_LOCK_EXIT(as);
5776 
5777 		error = uiomove((caddr_t)dirent, reclen, UIO_READ, uiop);
5778 		if (as != NULL)
5779 			AS_LOCK_ENTER(as, RW_WRITER);
5780 
5781 		if (error)
5782 			break;
5783 	}
5784 
5785 	if (error == 0 && eofp)
5786 		*eofp = (uiop->uio_offset >= (fddirsize + 2) * PRSDSIZE);
5787 
5788 	if (fip != NULL)
5789 		mutex_exit(&fip->fi_lock);
5790 	if (as != NULL)
5791 		AS_LOCK_EXIT(as);
5792 	mutex_enter(&p->p_lock);
5793 	prunlock(pnp);
5794 	return (error);
5795 }
5796 
5797 static int
5798 pr_readdir_tmpldir(prnode_t *pnp, uio_t *uiop, int *eofp)
5799 {
5800 	proc_t *p;
5801 	int pslot, tslot;
5802 	gfs_readdir_state_t gstate;
5803 	int error, eof = 0;
5804 	offset_t n;
5805 
5806 	ASSERT(pnp->pr_type == PR_TMPLDIR);
5807 
5808 	if ((error = prlock(pnp, ZNO)) != 0)
5809 		return (error);
5810 	p = pnp->pr_common->prc_proc;
5811 	pslot = pnp->pr_common->prc_slot;
5812 	tslot = pnp->pr_common->prc_tslot;
5813 	mutex_exit(&p->p_lock);
5814 
5815 	if ((error = gfs_readdir_init(&gstate, PRDIRSIZE, PRSDSIZE, uiop,
5816 	    pmkino(tslot, pslot, PR_LWPDIR),
5817 	    pmkino(tslot, pslot, PR_TMPLDIR), 0)) != 0) {
5818 		mutex_enter(&p->p_lock);
5819 		prunlock(pnp);
5820 		return (error);
5821 	}
5822 
5823 	while ((error = gfs_readdir_pred(&gstate, uiop, &n)) == 0) {
5824 		/*
5825 		 * Check for an active template.  Reading a directory's
5826 		 * contents is already racy, so we don't bother taking
5827 		 * any locks.
5828 		 */
5829 		while (n < ct_ntypes &&
5830 		    pnp->pr_common->prc_thread->t_lwp->lwp_ct_active[n] == NULL)
5831 			n++;
5832 		/*
5833 		 * Stop when all types have been reported.
5834 		 */
5835 		if (n >= ct_ntypes) {
5836 			eof = 1;
5837 			break;
5838 		}
5839 		/*
5840 		 * The pmkino invocation below will need to be updated
5841 		 * when we create our fifth contract type.
5842 		 */
5843 		ASSERT(ct_ntypes <= 4);
5844 		error = gfs_readdir_emit(&gstate, uiop, n,
5845 		    pmkino((tslot << 2) | n, pslot, PR_TMPL),
5846 		    ct_types[n]->ct_type_name, 0);
5847 		if (error)
5848 			break;
5849 	}
5850 
5851 	mutex_enter(&p->p_lock);
5852 	prunlock(pnp);
5853 
5854 	return (gfs_readdir_fini(&gstate, error, eofp, eof));
5855 }
5856 
5857 static int
5858 pr_readdir_ctdir(prnode_t *pnp, uio_t *uiop, int *eofp)
5859 {
5860 	proc_t *p;
5861 	int pslot;
5862 	gfs_readdir_state_t gstate;
5863 	int error, eof = 0;
5864 	offset_t n;
5865 	uint64_t zid;
5866 
5867 	ASSERT(pnp->pr_type == PR_CTDIR);
5868 
5869 	if ((error = prlock(pnp, ZNO)) != 0)
5870 		return (error);
5871 	p = pnp->pr_common->prc_proc;
5872 	pslot = p->p_slot;
5873 	mutex_exit(&p->p_lock);
5874 
5875 	if ((error = gfs_readdir_init(&gstate, PRDIRSIZE, PRSDSIZE, uiop,
5876 	    pmkino(0, pslot, PR_PIDDIR), pmkino(0, pslot, PR_CTDIR), 0)) != 0) {
5877 		mutex_enter(&p->p_lock);
5878 		prunlock(pnp);
5879 		return (error);
5880 	}
5881 
5882 	zid = VTOZONE(pnp->pr_vnode)->zone_uniqid;
5883 	while ((error = gfs_readdir_pred(&gstate, uiop, &n)) == 0) {
5884 		id_t next = contract_plookup(p, n, zid);
5885 		if (next == -1) {
5886 			eof = 1;
5887 			break;
5888 		}
5889 		error = gfs_readdir_emitn(&gstate, uiop, next,
5890 		    pmkino(next, pslot, PR_CT), next);
5891 		if (error)
5892 			break;
5893 	}
5894 
5895 	mutex_enter(&p->p_lock);
5896 	prunlock(pnp);
5897 
5898 	return (gfs_readdir_fini(&gstate, error, eofp, eof));
5899 }
5900 
5901 /* ARGSUSED */
5902 static int
5903 prfsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct)
5904 {
5905 	return (0);
5906 }
5907 
5908 /*
5909  * Utility: remove a /proc vnode from a linked list, threaded through pr_next.
5910  */
5911 static void
5912 pr_list_unlink(vnode_t *pvp, vnode_t **listp)
5913 {
5914 	vnode_t *vp;
5915 	prnode_t *pnp;
5916 
5917 	while ((vp = *listp) != NULL) {
5918 		pnp = VTOP(vp);
5919 		if (vp == pvp) {
5920 			*listp = pnp->pr_next;
5921 			pnp->pr_next = NULL;
5922 			break;
5923 		}
5924 		listp = &pnp->pr_next;
5925 	}
5926 }
5927 
5928 /* ARGSUSED */
5929 static void
5930 prinactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
5931 {
5932 	prnode_t *pnp = VTOP(vp);
5933 	prnodetype_t type = pnp->pr_type;
5934 	proc_t *p;
5935 	vnode_t *dp;
5936 	vnode_t *ovp = NULL;
5937 	prnode_t *opnp = NULL;
5938 
5939 	switch (type) {
5940 	case PR_OBJECT:
5941 	case PR_FD:
5942 	case PR_FDINFO:
5943 	case PR_SELF:
5944 	case PR_PATH:
5945 		/* These are not linked into the usual lists */
5946 		ASSERT(vp->v_count == 1);
5947 		if ((dp = pnp->pr_parent) != NULL)
5948 			VN_RELE(dp);
5949 		prfreenode(pnp);
5950 		return;
5951 	default:
5952 		break;
5953 	}
5954 
5955 	mutex_enter(&pr_pidlock);
5956 	if (pnp->pr_pcommon == NULL)
5957 		p = NULL;
5958 	else if ((p = pnp->pr_pcommon->prc_proc) != NULL)
5959 		mutex_enter(&p->p_lock);
5960 	mutex_enter(&vp->v_lock);
5961 
5962 	if (type == PR_PROCDIR || vp->v_count > 1) {
5963 		VN_RELE_LOCKED(vp);
5964 		mutex_exit(&vp->v_lock);
5965 		if (p != NULL)
5966 			mutex_exit(&p->p_lock);
5967 		mutex_exit(&pr_pidlock);
5968 		return;
5969 	}
5970 
5971 	if ((dp = pnp->pr_parent) != NULL) {
5972 		prnode_t *dpnp;
5973 
5974 		switch (type) {
5975 		case PR_PIDFILE:
5976 		case PR_LWPIDFILE:
5977 		case PR_OPAGEDATA:
5978 			break;
5979 		default:
5980 			dpnp = VTOP(dp);
5981 			mutex_enter(&dpnp->pr_mutex);
5982 			if (dpnp->pr_files != NULL &&
5983 			    dpnp->pr_files[pnp->pr_index] == vp)
5984 				dpnp->pr_files[pnp->pr_index] = NULL;
5985 			mutex_exit(&dpnp->pr_mutex);
5986 			break;
5987 		}
5988 		pnp->pr_parent = NULL;
5989 	}
5990 
5991 	ASSERT(vp->v_count == 1);
5992 
5993 	/*
5994 	 * If we allocated an old /proc/pid node, free it too.
5995 	 */
5996 	if (pnp->pr_pidfile != NULL) {
5997 		ASSERT(type == PR_PIDDIR);
5998 		ovp = pnp->pr_pidfile;
5999 		opnp = VTOP(ovp);
6000 		ASSERT(opnp->pr_type == PR_PIDFILE);
6001 		pnp->pr_pidfile = NULL;
6002 	}
6003 
6004 	mutex_exit(&pr_pidlock);
6005 
6006 	if (p != NULL) {
6007 		/*
6008 		 * Remove the vnodes from the lists of
6009 		 * /proc vnodes for the process.
6010 		 */
6011 		int slot;
6012 
6013 		switch (type) {
6014 		case PR_PIDDIR:
6015 			pr_list_unlink(vp, &p->p_trace);
6016 			break;
6017 		case PR_LWPIDDIR:
6018 			if ((slot = pnp->pr_common->prc_tslot) != -1) {
6019 				lwpent_t *lep = p->p_lwpdir[slot].ld_entry;
6020 				pr_list_unlink(vp, &lep->le_trace);
6021 			}
6022 			break;
6023 		default:
6024 			pr_list_unlink(vp, &p->p_plist);
6025 			break;
6026 		}
6027 		if (ovp != NULL)
6028 			pr_list_unlink(ovp, &p->p_plist);
6029 		mutex_exit(&p->p_lock);
6030 	}
6031 
6032 	mutex_exit(&vp->v_lock);
6033 
6034 	if (type == PR_CT && pnp->pr_contract != NULL) {
6035 		contract_rele(pnp->pr_contract);
6036 		pnp->pr_contract = NULL;
6037 	}
6038 
6039 	if (opnp != NULL)
6040 		prfreenode(opnp);
6041 	prfreenode(pnp);
6042 	if (dp != NULL) {
6043 		VN_RELE(dp);
6044 	}
6045 }
6046 
6047 /* ARGSUSED */
6048 static int
6049 prseek(vnode_t *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
6050 {
6051 	return (0);
6052 }
6053 
6054 /*
6055  * We use the p_execdir member of proc_t to expand the %d token in core file
6056  * paths (the directory path for the executable that dumped core; see
6057  * coreadm(8) for details). We'd like gcore(1) to be able to expand %d in
6058  * the same way as core dumping from the kernel, but there's no convenient
6059  * and comprehensible way to export the path name for p_execdir. To solve
6060  * this, we try to find the actual path to the executable that was used. In
6061  * pr_lookup_pathdir(), we mark the a.out path name vnode with the PR_AOUT
6062  * flag, and use that here to indicate that more work is needed beyond the
6063  * call to vnodetopath().
6064  */
6065 static int
6066 prreadlink_lookup(prnode_t *pnp, char *buf, size_t size, cred_t *cr)
6067 {
6068 	proc_t *p;
6069 	vnode_t *vp, *execvp, *vrootp;
6070 	int ret;
6071 	size_t len;
6072 	dirent64_t *dp;
6073 	size_t dlen = DIRENT64_RECLEN(MAXPATHLEN);
6074 	char *dbuf;
6075 
6076 	p = curproc;
6077 	mutex_enter(&p->p_lock);
6078 	if ((vrootp = PTOU(p)->u_rdir) == NULL)
6079 		vrootp = rootdir;
6080 	VN_HOLD(vrootp);
6081 	mutex_exit(&p->p_lock);
6082 
6083 	ret = vnodetopath(vrootp, pnp->pr_realvp, buf, size, cr);
6084 
6085 	/*
6086 	 * If PR_AOUT isn't set, then we looked up the path for the vnode;
6087 	 * otherwise, we looked up the path for (what we believe to be) the
6088 	 * containing directory.
6089 	 */
6090 	if ((pnp->pr_flags & PR_AOUT) == 0) {
6091 		VN_RELE(vrootp);
6092 		return (ret);
6093 	}
6094 
6095 	/*
6096 	 * Fail if there's a problem locking the process. This will only
6097 	 * occur if the process is changing so the information we would
6098 	 * report would already be invalid.
6099 	 */
6100 	if (prlock(pnp, ZNO) != 0) {
6101 		VN_RELE(vrootp);
6102 		return (EIO);
6103 	}
6104 
6105 	p = pnp->pr_common->prc_proc;
6106 	mutex_exit(&p->p_lock);
6107 
6108 	execvp = p->p_exec;
6109 	VN_HOLD(execvp);
6110 
6111 	/*
6112 	 * If our initial lookup of the directory failed, fall back to
6113 	 * the path name information for p_exec.
6114 	 */
6115 	if (ret != 0) {
6116 		mutex_enter(&p->p_lock);
6117 		prunlock(pnp);
6118 		ret = vnodetopath(vrootp, execvp, buf, size, cr);
6119 		VN_RELE(execvp);
6120 		VN_RELE(vrootp);
6121 		return (ret);
6122 	}
6123 
6124 	len = strlen(buf);
6125 
6126 	/*
6127 	 * We use u_comm as a guess for the last component of the full
6128 	 * executable path name. If there isn't going to be enough space
6129 	 * we fall back to using the p_exec so that we can have _an_
6130 	 * answer even if it's not perfect.
6131 	 */
6132 	if (strlen(PTOU(p)->u_comm) + len + 1 < size) {
6133 		buf[len] = '/';
6134 		(void) strcpy(buf + len + 1, PTOU(p)->u_comm);
6135 		mutex_enter(&p->p_lock);
6136 		prunlock(pnp);
6137 
6138 		/*
6139 		 * Do a forward lookup of our u_comm guess.
6140 		 */
6141 		if (lookupnameat(buf + len + 1, UIO_SYSSPACE, FOLLOW, NULLVPP,
6142 		    &vp, pnp->pr_realvp) == 0) {
6143 			if (vn_compare(vp, execvp)) {
6144 				VN_RELE(vp);
6145 				VN_RELE(execvp);
6146 				VN_RELE(vrootp);
6147 				return (0);
6148 			}
6149 
6150 			VN_RELE(vp);
6151 		}
6152 	} else {
6153 		mutex_enter(&p->p_lock);
6154 		prunlock(pnp);
6155 	}
6156 
6157 	dbuf = kmem_alloc(dlen, KM_SLEEP);
6158 
6159 	/*
6160 	 * Try to find a matching vnode by iterating through the directory's
6161 	 * entries. If that fails, fall back to the path information for
6162 	 * p_exec.
6163 	 */
6164 	if ((ret = dirfindvp(vrootp, pnp->pr_realvp, execvp, cr, dbuf,
6165 	    dlen, &dp)) == 0 && strlen(dp->d_name) + len + 1 < size) {
6166 		buf[len] = '/';
6167 		(void) strcpy(buf + len + 1, dp->d_name);
6168 	} else {
6169 		ret = vnodetopath(vrootp, execvp, buf, size, cr);
6170 	}
6171 
6172 	kmem_free(dbuf, dlen);
6173 	VN_RELE(execvp);
6174 	VN_RELE(vrootp);
6175 
6176 	return (ret);
6177 }
6178 
6179 /* ARGSUSED */
6180 static int
6181 prreadlink(vnode_t *vp, uio_t *uiop, cred_t *cr, caller_context_t *ctp)
6182 {
6183 	prnode_t *pnp = VTOP(vp);
6184 	char *buf;
6185 	int ret = EINVAL;
6186 	char idbuf[16];
6187 	int length, rlength;
6188 	contract_t *ct;
6189 
6190 	switch (pnp->pr_type) {
6191 	case PR_SELF:
6192 		(void) snprintf(idbuf, sizeof (idbuf), "%d", curproc->p_pid);
6193 		ret = uiomove(idbuf, strlen(idbuf), UIO_READ, uiop);
6194 		break;
6195 	case PR_OBJECT:
6196 	case PR_FD:
6197 	case PR_CURDIR:
6198 	case PR_ROOTDIR:
6199 		if (pnp->pr_realvp->v_type == VDIR)
6200 			ret = 0;
6201 		break;
6202 	case PR_PATH:
6203 		buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6204 
6205 		if ((ret = prreadlink_lookup(pnp, buf, MAXPATHLEN, cr)) == 0)
6206 			ret = uiomove(buf, strlen(buf), UIO_READ, uiop);
6207 
6208 		kmem_free(buf, MAXPATHLEN);
6209 		break;
6210 	case PR_CT:
6211 		ASSERT(pnp->pr_contract != NULL);
6212 		ct = pnp->pr_contract;
6213 		length = sizeof (CTFS_ROOT "//") + sizeof (idbuf) +
6214 		    strlen(ct->ct_type->ct_type_name);
6215 		buf = kmem_alloc(length, KM_SLEEP);
6216 		rlength = snprintf(buf, length, CTFS_ROOT "/%s/%d",
6217 		    ct->ct_type->ct_type_name, ct->ct_id);
6218 		ASSERT(rlength < length);
6219 		ret = uiomove(buf, rlength, UIO_READ, uiop);
6220 		kmem_free(buf, length);
6221 		break;
6222 	default:
6223 		break;
6224 	}
6225 
6226 	return (ret);
6227 }
6228 
6229 /*ARGSUSED2*/
6230 static int
6231 prcmp(vnode_t *vp1, vnode_t *vp2, caller_context_t *ct)
6232 {
6233 	prnode_t *pp1, *pp2;
6234 
6235 	if (vp1 == vp2)
6236 		return (1);
6237 
6238 	if (!vn_matchops(vp1, prvnodeops) || !vn_matchops(vp2, prvnodeops))
6239 		return (0);
6240 
6241 	pp1 = VTOP(vp1);
6242 	pp2 = VTOP(vp2);
6243 
6244 	if (pp1->pr_type != pp2->pr_type)
6245 		return (0);
6246 	if (pp1->pr_type == PR_PROCDIR)
6247 		return (1);
6248 	if (pp1->pr_ino || pp2->pr_ino)
6249 		return (pp2->pr_ino == pp1->pr_ino);
6250 
6251 	if (pp1->pr_common == NULL || pp2->pr_common == NULL)
6252 		return (0);
6253 
6254 	return (pp1->pr_common->prc_slot == pp2->pr_common->prc_slot &&
6255 	    pp1->pr_common->prc_tslot == pp2->pr_common->prc_tslot);
6256 }
6257 
6258 static int
6259 prrealvp(vnode_t *vp, vnode_t **vpp, caller_context_t *ct)
6260 {
6261 	vnode_t *rvp;
6262 
6263 	if ((rvp = VTOP(vp)->pr_realvp) != NULL) {
6264 		vp = rvp;
6265 		if (VOP_REALVP(vp, &rvp, ct) == 0)
6266 			vp = rvp;
6267 	}
6268 
6269 	*vpp = vp;
6270 	return (0);
6271 }
6272 
6273 /*
6274  * Return the answer requested to poll().
6275  * POLLIN, POLLRDNORM, and POLLOUT are recognized as in fs_poll().
6276  * In addition, these have special meaning for /proc files:
6277  *	POLLPRI		process or lwp stopped on an event of interest
6278  *	POLLERR		/proc file descriptor is invalid
6279  *	POLLHUP		process or lwp has terminated
6280  */
6281 /*ARGSUSED5*/
6282 static int
6283 prpoll(vnode_t *vp, short events, int anyyet, short *reventsp,
6284     pollhead_t **phpp, caller_context_t *ct)
6285 {
6286 	prnode_t *pnp = VTOP(vp);
6287 	prcommon_t *pcp = pnp->pr_common;
6288 	pollhead_t *php = &pcp->prc_pollhead;
6289 	proc_t *p;
6290 	short revents;
6291 	int error;
6292 	int lockstate;
6293 
6294 	ASSERT(pnp->pr_type < PR_NFILES);
6295 
6296 	/*
6297 	 * Support for old /proc interface.
6298 	 */
6299 	if (pnp->pr_pidfile != NULL) {
6300 		vp = pnp->pr_pidfile;
6301 		pnp = VTOP(vp);
6302 		ASSERT(pnp->pr_type == PR_PIDFILE);
6303 		ASSERT(pnp->pr_common == pcp);
6304 	}
6305 
6306 	*reventsp = revents = 0;
6307 	*phpp = (pollhead_t *)NULL;
6308 
6309 	if (vp->v_type == VDIR) {
6310 		*reventsp |= POLLNVAL;
6311 		return (0);
6312 	}
6313 
6314 	/* avoid deadlock with prnotify() */
6315 	if (pollunlock(&lockstate) != 0) {
6316 		*reventsp = POLLNVAL;
6317 		return (0);
6318 	}
6319 
6320 	if ((error = prlock(pnp, ZNO)) != 0) {
6321 		pollrelock(lockstate);
6322 		switch (error) {
6323 		case ENOENT:		/* process or lwp died */
6324 			*reventsp = POLLHUP;
6325 			error = 0;
6326 			break;
6327 		case EAGAIN:		/* invalidated */
6328 			*reventsp = POLLERR;
6329 			error = 0;
6330 			break;
6331 		}
6332 		return (error);
6333 	}
6334 
6335 	/*
6336 	 * We have the process marked locked (P_PR_LOCK) and we are holding
6337 	 * its p->p_lock.  We want to unmark the process but retain
6338 	 * exclusive control w.r.t. other /proc controlling processes
6339 	 * before reacquiring the polling locks.
6340 	 *
6341 	 * prunmark() does this for us.  It unmarks the process
6342 	 * but retains p->p_lock so we still have exclusive control.
6343 	 * We will drop p->p_lock at the end to relinquish control.
6344 	 *
6345 	 * We cannot call prunlock() at the end to relinquish control
6346 	 * because prunlock(), like prunmark(), may drop and reacquire
6347 	 * p->p_lock and that would lead to a lock order violation
6348 	 * w.r.t. the polling locks we are about to reacquire.
6349 	 */
6350 	p = pcp->prc_proc;
6351 	ASSERT(p != NULL);
6352 	prunmark(p);
6353 
6354 	pollrelock(lockstate);		/* reacquire dropped poll locks */
6355 
6356 	if ((p->p_flag & SSYS) || p->p_as == &kas)
6357 		revents = POLLNVAL;
6358 	else {
6359 		short ev;
6360 
6361 		if ((ev = (events & (POLLIN|POLLRDNORM))) != 0)
6362 			revents |= ev;
6363 		/*
6364 		 * POLLWRNORM (same as POLLOUT) really should not be
6365 		 * used to indicate that the process or lwp stopped.
6366 		 * However, USL chose to use POLLWRNORM rather than
6367 		 * POLLPRI to indicate this, so we just accept either
6368 		 * requested event to indicate stopped.  (grr...)
6369 		 */
6370 		if ((ev = (events & (POLLPRI|POLLOUT|POLLWRNORM))) != 0) {
6371 			kthread_t *t;
6372 
6373 			if (pcp->prc_flags & PRC_LWP) {
6374 				t = pcp->prc_thread;
6375 				ASSERT(t != NULL);
6376 				thread_lock(t);
6377 			} else {
6378 				t = prchoose(p);	/* returns locked t */
6379 				ASSERT(t != NULL);
6380 			}
6381 
6382 			if (ISTOPPED(t) || VSTOPPED(t))
6383 				revents |= ev;
6384 			thread_unlock(t);
6385 		}
6386 	}
6387 
6388 	*reventsp = revents;
6389 	if ((!anyyet && revents == 0) || (events & POLLET)) {
6390 		/*
6391 		 * Arrange to wake up the polling lwp when
6392 		 * the target process/lwp stops or terminates
6393 		 * or when the file descriptor becomes invalid.
6394 		 */
6395 		pcp->prc_flags |= PRC_POLL;
6396 		*phpp = php;
6397 	}
6398 	mutex_exit(&p->p_lock);
6399 	return (0);
6400 }
6401 
6402 /* in prioctl.c */
6403 extern int prioctl(vnode_t *, int, intptr_t, int, cred_t *, int *,
6404 	caller_context_t *);
6405 
6406 /*
6407  * /proc vnode operations vector
6408  */
6409 const fs_operation_def_t pr_vnodeops_template[] = {
6410 	VOPNAME_OPEN,		{ .vop_open = propen },
6411 	VOPNAME_CLOSE,		{ .vop_close = prclose },
6412 	VOPNAME_READ,		{ .vop_read = prread },
6413 	VOPNAME_WRITE,		{ .vop_write = prwrite },
6414 	VOPNAME_IOCTL,		{ .vop_ioctl = prioctl },
6415 	VOPNAME_GETATTR,	{ .vop_getattr = prgetattr },
6416 	VOPNAME_ACCESS,		{ .vop_access = praccess },
6417 	VOPNAME_LOOKUP,		{ .vop_lookup = prlookup },
6418 	VOPNAME_CREATE,		{ .vop_create = prcreate },
6419 	VOPNAME_READDIR,	{ .vop_readdir = prreaddir },
6420 	VOPNAME_READLINK,	{ .vop_readlink = prreadlink },
6421 	VOPNAME_FSYNC,		{ .vop_fsync = prfsync },
6422 	VOPNAME_INACTIVE,	{ .vop_inactive = prinactive },
6423 	VOPNAME_SEEK,		{ .vop_seek = prseek },
6424 	VOPNAME_CMP,		{ .vop_cmp = prcmp },
6425 	VOPNAME_FRLOCK,		{ .error = fs_error },
6426 	VOPNAME_REALVP,		{ .vop_realvp = prrealvp },
6427 	VOPNAME_POLL,		{ .vop_poll = prpoll },
6428 	VOPNAME_DISPOSE,	{ .error = fs_error },
6429 	VOPNAME_SHRLOCK,	{ .error = fs_error },
6430 	NULL,			NULL
6431 };
6432