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  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * File Events Notification
30  * ------------------------
31  *
32  * The File Events Notification facility provides file and directory change
33  * notification. It is implemented as an event source(PORT_SOURCE_FILE)
34  * under the Event Ports framework. Therefore the API is an extension to
35  * the Event Ports API.
36  *
37  * It uses the FEM (File Events Monitoring) framework to intercept
38  * operations on the files & directories and generate appropriate events.
39  *
40  * It provides event notification in accordance with what an application
41  * can find out by stat`ing the file and comparing time stamps. The various
42  * system calls that update the file's access, modification, and change
43  * time stamps are documented in the man page section 2.
44  *
45  * It is non intrusive. That is, having an active file event watch on a file
46  * or directory will not prevent it from being removed or renamed or block an
47  * unmount operation of the file system where the watched file or directory
48  * resides.
49  *
50  *
51  * Interface:
52  * ----------
53  *
54  *   The object for this event source is of type 'struct file_obj *'
55  *
56  *   The file that needs to be monitored is specified in 'fo_name'.
57  *   The time stamps collected by a stat(2) call are passed in fo_atime,
58  *   fo_mtime, fo_ctime. At the time a file events watch is registered, the
59  *   time stamps passed in are compared with the current time stamps of the
60  *   file. If it has changed, relevant events are sent immediately. If the time
61  *   stamps are all '0', they will not be compared.
62  *
63  *
64  * The events are delivered to an event port. A port is created using
65  * port_create().
66  *
67  * To register a file events watch on a file or directory.
68  *
69  *   port_associate(int port, PORT_SOURCE_FILE, (uintptr_t)&fobj, events, user)
70  *
71  *   'user' is the user pointer to be returned with the event.
72  *
73  * To de-register a file events watch,
74  *
75  *   port_dissociate(int port, PORT_SOURCE_FILE, (uintptr_t)&fobj)
76  *
77  * The events are collected using the port_get()/port_getn() interface. The
78  * event source will be PORT_SOURCE_FILE.
79  *
80  * After an event is delivered, the file events watch gets de-activated. To
81  * receive the next event, the process will have to re-register the watch and
82  * activate it by calling port_associate() again. This behavior is intentional
83  * and supports proper multi threaded programming when using file events
84  * notification API.
85  *
86  *
87  * Implementation overview:
88  * ------------------------
89  *
90  * Each file events watch is represented by 'portfop_t' in the kernel. A
91  * cache(in portfop_cache_t) of these portfop_t's are maintained per event
92  * port by this source. The object here is the pointer to the file_obj
93  * structure. The portfop_t's are hashed in using the object pointer. Therefore
94  * it is possible to have multiple file events watches on a file by the same
95  * process by using different object structure(file_obj_t) and hence can
96  * receive multiple event notification for a file. These watches can be for
97  * different event types.
98  *
99  * The cached entries of these file objects are retained, even after delivering
100  * an event, marking them inactive for performance reasons. The assumption
101  * is that the process would come back and re-register the file to receive
102  * further events. When there are more then 'port_fop_maxpfps' watches per file
103  * it will attempt to free the oldest inactive watches.
104  *
105  * In case the event that is being delivered is an exception event, the cached
106  * entries get removed. An exception event on a file or directory means its
107  * identity got changed(rename to/from, delete, mounted over, file system
108  * unmount).
109  *
110  * If the event port gets closed, all the associated file event watches will be
111  * removed and discarded.
112  *
113  *
114  * Data structures:
115  * ----------------
116  *
117  * The list of file event watches per file are managed by the data structure
118  * portfop_vp_t. The first time a file events watch is registered for a file,
119  * a portfop_vp_t is installed on the vnode_t's member v_fopdata. This gets
120  * removed and freed only when the vnode becomes inactive. The FEM hooks are
121  * also installed when the first watch is registered on a file. The FEM hooks
122  * get un-installed when all the watches are removed.
123  *
124  * Each file events watch is represented by the structure portfop_t. They
125  * get added to a list of portfop_t's on the vnode(portfop_vp_t). After
126  * delivering an event, the portfop_t is marked inactive but retained. It is
127  * moved to the end of the list. All the active portfop_t's are maintained at
128  * the beginning. In case of exception events, the portfop_t will be removed
129  * and discarded.
130  *
131  * To intercept unmount operations, FSEM hooks are added to the file system
132  * under which files are being watched. A hash table('portfop_vfs_hash_t') of
133  * active file systems is maintained. Each file system that has active watches
134  * is represented by 'portfop_vfs_t' and is added to the hash table.
135  * The vnode's 'portfop_vp_t' structure is added to the list of files(vnodes)
136  * being watched on the portfop_vfs_t structure.
137  *
138  *
139  * File system support:
140  * -------------------
141  *
142  * The file system implementation has to provide vnode event notifications
143  * (vnevents) in order to support watching any files on that file system.
144  * The vnode events(vnevents) are notifications provided by the file system
145  * for name based file operations like rename, remove etc, which do not go
146  * thru the VOP_** interfaces. If the file system does not implement vnode
147  * notifications, watching for file events on such file systems is not
148  * supported. The vnode event notifications support is determined by the call
149  * vnevent_support(vp) (VOP_VNEVENT(vp, VE_SUPPORT)), which the file system
150  * has to implement.
151  *
152  *
153  * Locking order:
154  * --------------
155  *
156  * A file(vnode) can have file event watches registered by different processes.
157  * There is one portfop_t per watch registered. These are on the vnode's list
158  * protected by the mutex 'pvp_mutex' in 'portfop_vp_t'. The portfop_t's are
159  * also on the per port cache. The cache is protected by the pfc_lock of
160  * portfop_cache_t. The lock order here is 'pfc_lock' -> 'pvp_mutex'.
161  *
162  */
163 
164 #include <sys/types.h>
165 #include <sys/systm.h>
166 #include <sys/stat.h>
167 #include <sys/errno.h>
168 #include <sys/kmem.h>
169 #include <sys/sysmacros.h>
170 #include <sys/debug.h>
171 #include <sys/vnode.h>
172 #include <sys/poll_impl.h>
173 #include <sys/port_impl.h>
174 #include <sys/fem.h>
175 #include <sys/vfs_opreg.h>
176 #include <sys/atomic.h>
177 #include <sys/mount.h>
178 #include <sys/mntent.h>
179 
180 /*
181  * For special case support of mnttab (/etc/mnttab).
182  */
183 extern struct vnode *vfs_mntdummyvp;
184 extern int mntfstype;
185 
186 #define	PORTFOP_PVFSH(vfsp)	(&portvfs_hash[PORTFOP_PVFSHASH(vfsp)])
187 portfop_vfs_hash_t	 portvfs_hash[PORTFOP_PVFSHASH_SZ];
188 
189 #define	PORTFOP_NVP	20
190 /*
191  * Inactive file event watches(portfop_t) are retained on the vnode's list
192  * for performance reason. If the applications re-registers the file, the
193  * inactive entry is made active and moved up the list.
194  *
195  * If there are greater then the following number of watches on a vnode,
196  * it will attempt to discard an oldest inactive watch(pfp) at the time
197  * a new watch is being registered and when events get delivered. We
198  * do this to avoid accumulating inactive watches on a file.
199  */
200 int	port_fop_maxpfps = 20;
201 
202 /* local functions */
203 static int	port_fop_callback(void *, int *, pid_t, int, void *);
204 
205 static void	port_pcache_insert(portfop_cache_t *, portfop_t *);
206 static void	port_pcache_delete(portfop_cache_t *, portfop_t *);
207 static void	port_close_fop(void *arg, int port, pid_t pid, int lastclose);
208 
209 /*
210  * port fop functions that will be the fem hooks.
211  */
212 static int port_fop_open(femarg_t *vf, int mode, cred_t *cr,
213     caller_context_t *);
214 static int port_fop_read(femarg_t *vf, uio_t *uiop, int ioflag, cred_t *cr,
215     struct caller_context *ct);
216 static int port_fop_write(femarg_t *vf, uio_t *uiop, int ioflag, cred_t *cr,
217     caller_context_t *ct);
218 static int port_fop_map(femarg_t *vf, offset_t off, struct as *as,
219     caddr_t *addrp, size_t len, uchar_t prot, uchar_t maxport,
220     uint_t flags, cred_t *cr, caller_context_t *ct);
221 static int port_fop_setattr(femarg_t *vf, vattr_t *vap, int flags, cred_t *cr,
222     caller_context_t *ct);
223 static int port_fop_create(femarg_t *vf, char *name, vattr_t *vap,
224     vcexcl_t excl, int mode, vnode_t **vpp, cred_t *cr, int flag,
225     caller_context_t *ct, vsecattr_t *vsecp);
226 static int port_fop_remove(femarg_t *vf, char *nm, cred_t *cr,
227     caller_context_t *ct, int flags);
228 static int port_fop_link(femarg_t *vf, vnode_t *svp, char *tnm, cred_t *cr,
229     caller_context_t *ct, int flags);
230 static int port_fop_rename(femarg_t *vf, char *snm, vnode_t *tdvp, char *tnm,
231     cred_t *cr, caller_context_t *ct, int flags);
232 static int port_fop_mkdir(femarg_t *vf, char *dirname, vattr_t *vap,
233     vnode_t **vpp, cred_t *cr, caller_context_t *ct, int flags,
234     vsecattr_t *vsecp);
235 static int port_fop_rmdir(femarg_t *vf, char *nm, vnode_t *cdir, cred_t *cr,
236     caller_context_t *ct, int flags);
237 static int port_fop_readdir(femarg_t *vf, uio_t *uiop, cred_t *cr, int *eofp,
238     caller_context_t *ct, int flags);
239 static int port_fop_symlink(femarg_t *vf, char *linkname, vattr_t *vap,
240     char *target, cred_t *cr, caller_context_t *ct, int flags);
241 static int port_fop_setsecattr(femarg_t *vf, vsecattr_t *vsap, int flag,
242     cred_t *cr, caller_context_t *ct);
243 
244 static int port_fop_vnevent(femarg_t *vf, vnevent_t vnevent, vnode_t *dvp,
245     char *cname, caller_context_t *ct);
246 
247 static int port_fop_unmount(fsemarg_t *vf, int flag, cred_t *cr);
248 
249 
250 /*
251  * Fem hooks.
252  */
253 const fs_operation_def_t	port_vnodesrc_template[] = {
254 	VOPNAME_OPEN,		{ .femop_open = port_fop_open },
255 	VOPNAME_READ,		{ .femop_read = port_fop_read },
256 	VOPNAME_WRITE,		{ .femop_write = port_fop_write },
257 	VOPNAME_MAP,		{ .femop_map = port_fop_map },
258 	VOPNAME_SETATTR, 	{ .femop_setattr = port_fop_setattr },
259 	VOPNAME_CREATE,		{ .femop_create = port_fop_create },
260 	VOPNAME_REMOVE,		{ .femop_remove = port_fop_remove },
261 	VOPNAME_LINK,		{ .femop_link = port_fop_link },
262 	VOPNAME_RENAME,		{ .femop_rename = port_fop_rename },
263 	VOPNAME_MKDIR,		{ .femop_mkdir = port_fop_mkdir },
264 	VOPNAME_RMDIR,		{ .femop_rmdir = port_fop_rmdir },
265 	VOPNAME_READDIR,	{ .femop_readdir = port_fop_readdir },
266 	VOPNAME_SYMLINK,	{ .femop_symlink = port_fop_symlink },
267 	VOPNAME_SETSECATTR, 	{ .femop_setsecattr = port_fop_setsecattr },
268 	VOPNAME_VNEVENT,	{ .femop_vnevent = port_fop_vnevent },
269 	NULL,	NULL
270 };
271 
272 /*
273  * Fsem - vfs ops hooks
274  */
275 const fs_operation_def_t	port_vfssrc_template[] = {
276 	VFSNAME_UNMOUNT, 	{ .fsemop_unmount = port_fop_unmount },
277 	NULL,	NULL
278 };
279 
280 fem_t *fop_femop;
281 fsem_t *fop_fsemop;
282 
283 static fem_t *
284 port_fop_femop()
285 {
286 	fem_t *femp;
287 	if (fop_femop != NULL)
288 		return (fop_femop);
289 	if (fem_create("portfop_fem",
290 	    (const struct fs_operation_def *)port_vnodesrc_template,
291 	    (fem_t **)&femp)) {
292 		return (NULL);
293 	}
294 	if (casptr(&fop_femop, NULL, femp) != NULL) {
295 		/*
296 		 * some other thread beat us to it.
297 		 */
298 		fem_free(femp);
299 	}
300 	return (fop_femop);
301 }
302 
303 static fsem_t *
304 port_fop_fsemop()
305 {
306 	fsem_t *fsemp;
307 	if (fop_fsemop != NULL)
308 		return (fop_fsemop);
309 	if (fsem_create("portfop_fsem", port_vfssrc_template, &fsemp)) {
310 		return (NULL);
311 	}
312 	if (casptr(&fop_fsemop, NULL, fsemp) != NULL) {
313 		/*
314 		 * some other thread beat us to it.
315 		 */
316 		fsem_free(fsemp);
317 	}
318 	return (fop_fsemop);
319 }
320 
321 /*
322  * port_fop_callback()
323  * - PORT_CALLBACK_DEFAULT
324  *	The file event will be delivered to the application.
325  * - PORT_CALLBACK_DISSOCIATE
326  *	The object will be dissociated from  the port.
327  * - PORT_CALLBACK_CLOSE
328  *	The object will be dissociated from the port because the port
329  *	is being closed.
330  */
331 /* ARGSUSED */
332 static int
333 port_fop_callback(void *arg, int *events, pid_t pid, int flag, void *evp)
334 {
335 	portfop_t	*pfp = (portfop_t *)arg;
336 	port_kevent_t	*pkevp = (port_kevent_t *)evp;
337 	int		error = 0;
338 
339 	ASSERT((events != NULL));
340 	if (flag == PORT_CALLBACK_DEFAULT) {
341 		if (curproc->p_pid != pid) {
342 				return (EACCES); /* deny delivery of events */
343 		}
344 
345 		*events = pkevp->portkev_events;
346 		pkevp->portkev_events = 0;
347 		if (pfp != NULL) {
348 			pfp->pfop_flags &= ~PORT_FOP_KEV_ONQ;
349 		}
350 	}
351 	return (error);
352 }
353 
354 /*
355  * Inserts a portfop_t into the port sources cache's.
356  */
357 static void
358 port_pcache_insert(portfop_cache_t *pfcp, portfop_t *pfp)
359 {
360 	portfop_t	**bucket;
361 
362 	ASSERT(MUTEX_HELD(&pfcp->pfc_lock));
363 	bucket = PORT_FOP_BUCKET(pfcp, pfp->pfop_object);
364 	pfp->pfop_hashnext = *bucket;
365 	*bucket = pfp;
366 	pfcp->pfc_objcount++;
367 }
368 
369 /*
370  * Remove the pfp from the port source cache.
371  */
372 static void
373 port_pcache_delete(portfop_cache_t *pfcp, portfop_t *pfp)
374 {
375 	portfop_t	*lpdp;
376 	portfop_t	*cpdp;
377 	portfop_t	**bucket;
378 
379 	bucket = PORT_FOP_BUCKET(pfcp, pfp->pfop_object);
380 	cpdp = *bucket;
381 	if (pfp == cpdp) {
382 		*bucket = pfp->pfop_hashnext;
383 	} else {
384 		while (cpdp != NULL) {
385 			lpdp = cpdp;
386 			cpdp = cpdp->pfop_hashnext;
387 			if (cpdp == pfp) {
388 				/* portfop struct found */
389 				lpdp->pfop_hashnext = pfp->pfop_hashnext;
390 				break;
391 			}
392 		}
393 	}
394 	pfcp->pfc_objcount--;
395 }
396 
397 /*
398  * The vnode's(portfop_vp_t) pfp list management. The 'pvp_mutex' is held
399  * when these routines are called.
400  *
401  * The 'pvp_lpfop' member points to the oldest inactive entry on the list.
402  * It is used to discard the oldtest inactive pfp if the number of entries
403  * exceed the limit.
404  */
405 static void
406 port_fop_listinsert(portfop_vp_t *pvp, portfop_t *pfp, int where)
407 {
408 	if (where == 1) {
409 		list_insert_head(&pvp->pvp_pfoplist, (void *)pfp);
410 	} else {
411 		list_insert_tail(&pvp->pvp_pfoplist, (void *)pfp);
412 	}
413 	if (pvp->pvp_lpfop == NULL) {
414 		pvp->pvp_lpfop = pfp;
415 	}
416 	pvp->pvp_cnt++;
417 }
418 
419 static void
420 port_fop_listinsert_head(portfop_vp_t *pvp, portfop_t *pfp)
421 {
422 	port_fop_listinsert(pvp, pfp, 1);
423 }
424 
425 static void
426 port_fop_listinsert_tail(portfop_vp_t *pvp, portfop_t *pfp)
427 {
428 	/*
429 	 * We point lpfop to an inactive one, if it was initially pointing
430 	 * to an active one. Insert to the tail is done only when a pfp goes
431 	 * inactive.
432 	 */
433 	if (pvp->pvp_lpfop && pvp->pvp_lpfop->pfop_flags & PORT_FOP_ACTIVE) {
434 		pvp->pvp_lpfop = pfp;
435 	}
436 	port_fop_listinsert(pvp, pfp, 0);
437 }
438 
439 static void
440 port_fop_listremove(portfop_vp_t *pvp, portfop_t *pfp)
441 {
442 	if (pvp->pvp_lpfop == pfp) {
443 		pvp->pvp_lpfop = list_next(&pvp->pvp_pfoplist, (void *)pfp);
444 	}
445 
446 	list_remove(&pvp->pvp_pfoplist, (void *)pfp);
447 
448 	pvp->pvp_cnt--;
449 	if (pvp->pvp_cnt && pvp->pvp_lpfop == NULL) {
450 		pvp->pvp_lpfop = list_head(&pvp->pvp_pfoplist);
451 	}
452 }
453 
454 static void
455 port_fop_listmove(portfop_vp_t *pvp, list_t *tlist)
456 {
457 	list_move_tail(tlist, &pvp->pvp_pfoplist);
458 	pvp->pvp_lpfop = NULL;
459 	pvp->pvp_cnt = 0;
460 }
461 
462 /*
463  * Remove a portfop_t from the port cache hash table and discard it.
464  * It is called only when pfp is not on the vnode's list. Otherwise,
465  * port_remove_fop() is called.
466  */
467 void
468 port_pcache_remove_fop(portfop_cache_t *pfcp, portfop_t *pfp)
469 {
470 	port_kevent_t	*pkevp;
471 
472 
473 	ASSERT(MUTEX_HELD(&pfcp->pfc_lock));
474 
475 	pkevp = pfp->pfop_pev;
476 	pfp->pfop_pev = NULL;
477 
478 	if (pkevp != NULL) {
479 		(void) port_remove_done_event(pkevp);
480 		port_free_event_local(pkevp, 0);
481 	}
482 
483 	port_pcache_delete(pfcp, pfp);
484 
485 	if (pfp->pfop_cname != NULL)
486 		kmem_free(pfp->pfop_cname, pfp->pfop_clen + 1);
487 	kmem_free(pfp, sizeof (portfop_t));
488 	if (pfcp->pfc_objcount == 0)
489 		cv_signal(&pfcp->pfc_lclosecv);
490 }
491 
492 /*
493  * if we have too many watches on the vnode, attempt to discard an
494  * inactive one.
495  */
496 static void
497 port_fop_trimpfplist(vnode_t *vp)
498 {
499 	portfop_vp_t *pvp;
500 	portfop_t *pfp = NULL;
501 	portfop_cache_t *pfcp;
502 	vnode_t	*tdvp;
503 
504 	/*
505 	 * Due to a reference the vnode cannot disappear, v_fopdata should
506 	 * not change.
507 	 */
508 	if ((pvp = vp->v_fopdata) != NULL &&
509 	    pvp->pvp_cnt > port_fop_maxpfps) {
510 		mutex_enter(&pvp->pvp_mutex);
511 		pfp = pvp->pvp_lpfop;
512 		pfcp = pfp->pfop_pcache;
513 		/*
514 		 * only if we can get the cache lock, we need to
515 		 * do this due to reverse lock order and some thread
516 		 * that may be trying to reactivate this entry.
517 		 */
518 		if (mutex_tryenter(&pfcp->pfc_lock)) {
519 			if (pfp && !(pfp->pfop_flags & PORT_FOP_ACTIVE) &&
520 			    !(pfp->pfop_flags & PORT_FOP_KEV_ONQ)) {
521 				port_fop_listremove(pvp, pfp);
522 				pfp->pfop_flags |= PORT_FOP_REMOVING;
523 			} else {
524 				mutex_exit(&pfcp->pfc_lock);
525 				pfp = NULL;
526 			}
527 		} else {
528 			pfp = NULL;
529 		}
530 		mutex_exit(&pvp->pvp_mutex);
531 
532 		/*
533 		 * discard pfp if any.
534 		 */
535 		if (pfp != NULL) {
536 			tdvp = pfp->pfop_dvp;
537 			port_pcache_remove_fop(pfcp, pfp);
538 			mutex_exit(&pfcp->pfc_lock);
539 			if (tdvp != NULL)
540 				VN_RELE(tdvp);
541 		}
542 	}
543 }
544 
545 /*
546  * This routine returns 1, if the vnode can be rele'ed by the caller.
547  * The caller has to VN_RELE the vnode with out holding any
548  * locks.
549  */
550 int
551 port_fop_femuninstall(vnode_t *vp)
552 {
553 	portfop_vp_t	*pvp;
554 	vfs_t		*vfsp;
555 	portfop_vfs_t *pvfsp;
556 	portfop_vfs_hash_t	*pvfsh;
557 	kmutex_t	*mtx;
558 	int	ret = 0;
559 
560 	/*
561 	 * if list is empty, uninstall fem.
562 	 */
563 	pvp = vp->v_fopdata;
564 	ASSERT(MUTEX_HELD(&pvp->pvp_mutex));
565 
566 	/*
567 	 * make sure the list is empty.
568 	 */
569 	if (!list_head(&pvp->pvp_pfoplist)) {
570 
571 		/*
572 		 * we could possibly uninstall the fem hooks when
573 		 * the vnode becomes inactive and the v_fopdata is
574 		 * free. But the hooks get triggered unnecessarily
575 		 * even though there are no active watches. So, we
576 		 * uninstall it here.
577 		 */
578 		(void) fem_uninstall(vp, (fem_t *)pvp->pvp_femp, vp);
579 		pvp->pvp_femp = NULL;
580 		mutex_exit(&pvp->pvp_mutex);
581 
582 
583 		/*
584 		 * If we successfully uninstalled fem, no process is watching
585 		 * this vnode, Remove it from the vfs's list of watched vnodes.
586 		 */
587 		pvfsp = pvp->pvp_pvfsp;
588 		vfsp = vp->v_vfsp;
589 		pvfsh = PORTFOP_PVFSH(vfsp);
590 		mtx = &pvfsh->pvfshash_mutex;
591 		mutex_enter(mtx);
592 		/*
593 		 * If unmount is in progress, that thread will remove and
594 		 * release the vnode from the vfs's list, just leave.
595 		 */
596 		if (!pvfsp->pvfs_unmount) {
597 			list_remove(&pvfsp->pvfs_pvplist, pvp);
598 			mutex_exit(mtx);
599 			ret = 1;
600 		} else {
601 			mutex_exit(mtx);
602 		}
603 	} else {
604 		mutex_exit(&pvp->pvp_mutex);
605 	}
606 	return (ret);
607 }
608 
609 /*
610  * Remove pfp from the vnode's watch list and the cache and discard it.
611  * If it is the last pfp on the vnode's list, the fem hooks get uninstalled.
612  * Returns 1 if pfp removed successfully.
613  *
614  * The *active is set to indicate if the pfp was still active(no events had
615  * been posted, or the posted event had not been collected yet and it was
616  * able to remove it from the port's queue).
617  *
618  * vpp and dvpp will point to the vnode and directory vnode which the caller
619  * is required to VN_RELE without holding any locks.
620  */
621 int
622 port_remove_fop(portfop_t *pfp, portfop_cache_t *pfcp, int cleanup,
623     int *active, vnode_t **vpp, vnode_t **dvpp)
624 {
625 	vnode_t		*vp;
626 	portfop_vp_t	*pvp;
627 	int	tactive = 0;
628 
629 	ASSERT(MUTEX_HELD(&pfcp->pfc_lock));
630 	vp = pfp->pfop_vp;
631 	pvp = vp->v_fopdata;
632 	mutex_enter(&pvp->pvp_mutex);
633 
634 	/*
635 	 * if not cleanup, remove it only if the pfp is still active and
636 	 * is not being removed by some other thread.
637 	 */
638 	if (!cleanup && (!(pfp->pfop_flags & PORT_FOP_ACTIVE) ||
639 	    pfp->pfop_flags & PORT_FOP_REMOVING)) {
640 		mutex_exit(&pvp->pvp_mutex);
641 		return (0);
642 	}
643 
644 	/*
645 	 * mark it inactive.
646 	 */
647 	if (pfp->pfop_flags & PORT_FOP_ACTIVE) {
648 		pfp->pfop_flags &= ~PORT_FOP_ACTIVE;
649 		tactive = 1;
650 	}
651 
652 	/*
653 	 * Check if the pfp is still on the vnode's list. This can
654 	 * happen if port_fop_excep() is in the process of removing it.
655 	 * In case of cleanup, just mark this pfp as inactive so that no
656 	 * new events (VNEVENT) will be delivered, and remove it from the
657 	 * event queue if it was already queued. Since the cache lock is
658 	 * held, the pfp will not disappear, even though it is being
659 	 * removed.
660 	 */
661 	if (pfp->pfop_flags & PORT_FOP_REMOVING) {
662 		mutex_exit(&pvp->pvp_mutex);
663 		if (!tactive && port_remove_done_event(pfp->pfop_pev)) {
664 			pfp->pfop_flags &= ~PORT_FOP_KEV_ONQ;
665 			tactive = 1;
666 		}
667 		if (active) {
668 			*active = tactive;
669 		}
670 		return (1);
671 	}
672 
673 	/*
674 	 * if we find an event on the queue and removed it, then this
675 	 * association is considered active.
676 	 */
677 	if (!tactive && port_remove_done_event(pfp->pfop_pev)) {
678 		pfp->pfop_flags &= ~PORT_FOP_KEV_ONQ;
679 		tactive = 1;
680 	}
681 
682 	if (active) {
683 		*active = tactive;
684 	}
685 	pvp = (portfop_vp_t *)vp->v_fopdata;
686 
687 	/*
688 	 * remove pfp from the vnode's list
689 	 */
690 	port_fop_listremove(pvp, pfp);
691 
692 	/*
693 	 * If no more associations on the vnode, uninstall fem hooks.
694 	 * The pvp mutex will be released in this routine.
695 	 */
696 	if (port_fop_femuninstall(vp))
697 		*vpp = vp;
698 	*dvpp = pfp->pfop_dvp;
699 	port_pcache_remove_fop(pfcp, pfp);
700 	return (1);
701 }
702 
703 /*
704  * This routine returns a pointer to a cached portfop entry, or NULL if it
705  * does not find it in the hash table. The object pointer is used as index.
706  * The entries are hashed by the object's address. We need to match the pid
707  * as the evet port can be shared between processes. The file events
708  * watches are per process only.
709  */
710 portfop_t *
711 port_cache_lookup_fop(portfop_cache_t *pfcp, pid_t pid, uintptr_t obj)
712 {
713 	portfop_t	*pfp = NULL;
714 	portfop_t	**bucket;
715 
716 	ASSERT(MUTEX_HELD(&pfcp->pfc_lock));
717 	bucket = PORT_FOP_BUCKET(pfcp, obj);
718 	pfp = *bucket;
719 	while (pfp != NULL) {
720 		if (pfp->pfop_object == obj && pfp->pfop_pid == pid)
721 			break;
722 		pfp = pfp->pfop_hashnext;
723 	}
724 	return (pfp);
725 }
726 
727 /*
728  * Given the file name, get the vnode and also the directory vnode
729  * On return, the vnodes are held (VN_HOLD). The caller has to VN_RELE
730  * the vnode(s).
731  */
732 int
733 port_fop_getdvp(void *objptr, vnode_t **vp, vnode_t **dvp,
734 	char **cname, int *len, int follow)
735 {
736 	int error = 0;
737 	struct pathname pn;
738 	char *fname;
739 
740 	if (get_udatamodel() == DATAMODEL_NATIVE) {
741 		fname = ((file_obj_t *)objptr)->fo_name;
742 #ifdef  _SYSCALL32_IMPL
743 	} else {
744 		fname = (caddr_t)(uintptr_t)((file_obj32_t *)objptr)->fo_name;
745 #endif	/* _SYSCALL32_IMPL */
746 	}
747 
748 	/*
749 	 * lookuppn may fail with EINVAL, if dvp is  non-null(like when
750 	 * looking for "."). So call again with dvp = NULL.
751 	 */
752 	if ((error = pn_get(fname, UIO_USERSPACE, &pn)) != 0) {
753 		return (error);
754 	}
755 
756 	error = lookuppn(&pn, NULL, follow, dvp, vp);
757 	if (error == EINVAL) {
758 		pn_free(&pn);
759 		if ((error = pn_get(fname, UIO_USERSPACE, &pn)) != 0) {
760 			return (error);
761 		}
762 		error = lookuppn(&pn, NULL, follow, NULL, vp);
763 		if (dvp != NULL) {
764 			*dvp = NULL;
765 		}
766 	}
767 
768 	if (error == 0 && cname != NULL && len != NULL) {
769 		pn_setlast(&pn);
770 		*len = pn.pn_pathlen;
771 		*cname = kmem_alloc(*len + 1, KM_SLEEP);
772 		(void) strcpy(*cname, pn.pn_path);
773 	} else {
774 		if (cname != NULL && len != NULL) {
775 			*cname = NULL;
776 			*len = 0;
777 		}
778 	}
779 
780 	pn_free(&pn);
781 	return (error);
782 }
783 
784 port_source_t *
785 port_getsrc(port_t *pp, int source)
786 {
787 	port_source_t *pse;
788 	int	lock = 0;
789 	/*
790 	 * get the port source structure.
791 	 */
792 	if (!MUTEX_HELD(&pp->port_queue.portq_source_mutex)) {
793 		mutex_enter(&pp->port_queue.portq_source_mutex);
794 		lock = 1;
795 	}
796 
797 	pse = pp->port_queue.portq_scache[PORT_SHASH(source)];
798 	for (; pse != NULL; pse = pse->portsrc_next) {
799 		if (pse->portsrc_source == source)
800 			break;
801 	}
802 
803 	if (lock) {
804 		mutex_exit(&pp->port_queue.portq_source_mutex);
805 	}
806 	return (pse);
807 }
808 
809 
810 /*
811  * Compare time stamps and generate an event if it has changed.
812  * Note that the port cache pointer will be valid due to a reference
813  * to the port. We need to grab the port cache lock and verify that
814  * the pfp is still the same before proceeding to deliver an event.
815  */
816 static void
817 port_check_timestamp(portfop_cache_t *pfcp, vnode_t *vp, vnode_t *dvp,
818 	portfop_t *pfp, void *objptr, uintptr_t object)
819 {
820 	vattr_t		vatt;
821 	portfop_vp_t	*pvp = vp->v_fopdata;
822 	int		events = 0;
823 	port_kevent_t	*pkevp;
824 	file_obj_t	*fobj;
825 	portfop_t	*tpfp;
826 
827 	/*
828 	 * If time stamps are specified, get attributes and compare.
829 	 */
830 	vatt.va_mask = AT_ATIME|AT_MTIME|AT_CTIME;
831 	if (get_udatamodel() == DATAMODEL_NATIVE) {
832 		fobj = (file_obj_t *)objptr;
833 		if (fobj->fo_atime.tv_sec || fobj->fo_atime.tv_nsec ||
834 		    fobj->fo_mtime.tv_sec || fobj->fo_mtime.tv_nsec ||
835 		    fobj->fo_ctime.tv_sec || fobj->fo_ctime.tv_nsec) {
836 			if (VOP_GETATTR(vp, &vatt, 0, CRED(), NULL)) {
837 				return;
838 			}
839 		} else {
840 			/*
841 			 * timestamp not specified, all 0's,
842 			 */
843 			return;
844 		}
845 #ifdef  _SYSCALL32_IMPL
846 	} else {
847 		file_obj32_t	*fobj32;
848 		fobj32 = (file_obj32_t *)objptr;
849 		if (fobj32->fo_atime.tv_sec || fobj32->fo_atime.tv_nsec ||
850 		    fobj32->fo_mtime.tv_sec || fobj32->fo_mtime.tv_nsec ||
851 		    fobj32->fo_ctime.tv_sec || fobj32->fo_ctime.tv_nsec) {
852 			if (VOP_GETATTR(vp, &vatt, 0, CRED(), NULL)) {
853 				return;
854 			}
855 		} else {
856 			/*
857 			 * timestamp not specified, all 0.
858 			 */
859 			return;
860 		}
861 #endif /* _SYSCALL32_IMPL */
862 	}
863 
864 	/*
865 	 * Now grab the cache lock and verify that we are still
866 	 * dealing with the same pfp and curthread is the one
867 	 * which registered it. We need to do this to avoid
868 	 * delivering redundant events.
869 	 */
870 	mutex_enter(&pfcp->pfc_lock);
871 	tpfp = port_cache_lookup_fop(pfcp, curproc->p_pid, object);
872 
873 	if (tpfp == NULL || tpfp != pfp ||
874 	    pfp->pfop_vp != vp || pfp->pfop_dvp != dvp ||
875 	    pfp->pfop_callrid != curthread ||
876 	    !(pfp->pfop_flags & PORT_FOP_ACTIVE)) {
877 		/*
878 		 * Some other event was delivered, the file
879 		 * watch was removed or reassociated. Just
880 		 * ignore it and leave
881 		 */
882 		mutex_exit(&pfcp->pfc_lock);
883 		return;
884 	}
885 
886 	mutex_enter(&pvp->pvp_mutex);
887 	/*
888 	 * The pfp cannot disappear as the port cache lock is held.
889 	 * While the pvp_mutex is held, no events will get delivered.
890 	 */
891 	if (pfp->pfop_flags & PORT_FOP_ACTIVE &&
892 	    !(pfp->pfop_flags & PORT_FOP_REMOVING)) {
893 		if (get_udatamodel() == DATAMODEL_NATIVE) {
894 			fobj = (file_obj_t *)objptr;
895 			if (pfp->pfop_events & FILE_ACCESS &&
896 			    (fobj->fo_atime.tv_sec || fobj->fo_atime.tv_nsec) &&
897 			    (vatt.va_atime.tv_sec != fobj->fo_atime.tv_sec ||
898 			    vatt.va_atime.tv_nsec != fobj->fo_atime.tv_nsec))
899 				events |= FILE_ACCESS;
900 
901 			if (pfp->pfop_events & FILE_MODIFIED &&
902 			    (fobj->fo_mtime.tv_sec || fobj->fo_mtime.tv_nsec) &&
903 			    (vatt.va_mtime.tv_sec != fobj->fo_mtime.tv_sec ||
904 			    vatt.va_mtime.tv_nsec != fobj->fo_mtime.tv_nsec))
905 				events |= FILE_MODIFIED;
906 
907 			if (pfp->pfop_events & FILE_ATTRIB &&
908 			    (fobj->fo_ctime.tv_sec || fobj->fo_ctime.tv_nsec) &&
909 			    (vatt.va_ctime.tv_sec != fobj->fo_ctime.tv_sec ||
910 			    vatt.va_ctime.tv_nsec != fobj->fo_ctime.tv_nsec))
911 				events |= FILE_ATTRIB;
912 #ifdef  _SYSCALL32_IMPL
913 		} else {
914 			file_obj32_t	*fobj32;
915 			fobj32 = (file_obj32_t *)objptr;
916 			if (pfp->pfop_events & FILE_ACCESS &&
917 			    (fobj32->fo_atime.tv_sec ||
918 			    fobj32->fo_atime.tv_nsec) &&
919 			    (vatt.va_atime.tv_sec != fobj32->fo_atime.tv_sec ||
920 			    vatt.va_atime.tv_nsec != fobj32->fo_atime.tv_nsec))
921 				events |= FILE_ACCESS;
922 
923 			if (pfp->pfop_events & FILE_MODIFIED &&
924 			    (fobj32->fo_mtime.tv_sec ||
925 			    fobj32->fo_mtime.tv_nsec) &&
926 			    (vatt.va_mtime.tv_sec != fobj32->fo_mtime.tv_sec ||
927 			    vatt.va_mtime.tv_nsec != fobj32->fo_mtime.tv_nsec))
928 				events |= FILE_MODIFIED;
929 
930 			if (pfp->pfop_events & FILE_ATTRIB &&
931 			    (fobj32->fo_ctime.tv_sec ||
932 			    fobj32->fo_ctime.tv_nsec) &&
933 			    (vatt.va_ctime.tv_sec != fobj32->fo_ctime.tv_sec ||
934 			    vatt.va_ctime.tv_nsec != fobj32->fo_ctime.tv_nsec))
935 				events |= FILE_ATTRIB;
936 #endif /* _SYSCALL32_IMPL */
937 		}
938 
939 		/*
940 		 * No events to deliver
941 		 */
942 		if (events == 0) {
943 			mutex_exit(&pvp->pvp_mutex);
944 			mutex_exit(&pfcp->pfc_lock);
945 			return;
946 		}
947 
948 		/*
949 		 * Deliver the event now.
950 		 */
951 		pkevp = pfp->pfop_pev;
952 		pfp->pfop_flags &= ~PORT_FOP_ACTIVE;
953 		pkevp->portkev_events |= events;
954 		/*
955 		 * Move it to the tail as active once are in the
956 		 * beginning of the list.
957 		 */
958 		port_fop_listremove(pvp, pfp);
959 		port_fop_listinsert_tail(pvp, pfp);
960 		port_send_event(pkevp);
961 		pfp->pfop_flags |= PORT_FOP_KEV_ONQ;
962 	}
963 	mutex_exit(&pvp->pvp_mutex);
964 	mutex_exit(&pfcp->pfc_lock);
965 }
966 
967 /*
968  * Add the event source to the port and return the port source cache pointer.
969  */
970 int
971 port_fop_associate_source(portfop_cache_t **pfcpp, port_t *pp, int source)
972 {
973 	portfop_cache_t *pfcp;
974 	port_source_t	*pse;
975 	int		error;
976 
977 	/*
978 	 * associate PORT_SOURCE_FILE source with the port, if it is
979 	 * not associated yet. Note the PORT_SOURCE_FILE source is
980 	 * associated once and will not be dissociated.
981 	 */
982 	if ((pse = port_getsrc(pp, PORT_SOURCE_FILE)) == NULL) {
983 		if (error = port_associate_ksource(pp->port_fd, source,
984 		    &pse, port_close_fop, pp, NULL)) {
985 			*pfcpp = NULL;
986 			return (error);
987 		}
988 	}
989 
990 	/*
991 	 * Get the portfop cache pointer.
992 	 */
993 	if ((pfcp = pse->portsrc_data) == NULL) {
994 		/*
995 		 * This is the first time that a file is being associated,
996 		 * create the portfop cache.
997 		 */
998 		pfcp = kmem_zalloc(sizeof (portfop_cache_t), KM_SLEEP);
999 		mutex_enter(&pp->port_queue.portq_source_mutex);
1000 		if (pse->portsrc_data == NULL) {
1001 			pse->portsrc_data = pfcp;
1002 			mutex_exit(&pp->port_queue.portq_source_mutex);
1003 		} else {
1004 			/*
1005 			 * someone else created the port cache, free
1006 			 * what we just now allocated.
1007 			 */
1008 			mutex_exit(&pp->port_queue.portq_source_mutex);
1009 			kmem_free(pfcp, sizeof (portfop_cache_t));
1010 			pfcp = pse->portsrc_data;
1011 		}
1012 	}
1013 	*pfcpp = pfcp;
1014 	return (0);
1015 }
1016 
1017 /*
1018  * Add the given pvp on the file system's list of vnodes watched.
1019  */
1020 int
1021 port_fop_pvfsadd(portfop_vp_t *pvp)
1022 {
1023 	int error = 0;
1024 	vnode_t	*vp = pvp->pvp_vp;
1025 	portfop_vfs_hash_t *pvfsh;
1026 	portfop_vfs_t	 *pvfsp;
1027 	fsem_t		*fsemp;
1028 
1029 	pvfsh = PORTFOP_PVFSH(vp->v_vfsp);
1030 	mutex_enter(&pvfsh->pvfshash_mutex);
1031 	for (pvfsp = pvfsh->pvfshash_pvfsp; pvfsp &&
1032 	    pvfsp->pvfs != vp->v_vfsp; pvfsp = pvfsp->pvfs_next)
1033 		;
1034 
1035 	if (!pvfsp) {
1036 		if ((fsemp = port_fop_fsemop()) != NULL) {
1037 			if ((error = fsem_install(vp->v_vfsp, fsemp,
1038 			    vp->v_vfsp, OPUNIQ, NULL, NULL))) {
1039 				mutex_exit(&pvfsh->pvfshash_mutex);
1040 				return (error);
1041 			}
1042 		} else {
1043 			mutex_exit(&pvfsh->pvfshash_mutex);
1044 			return (EINVAL);
1045 		}
1046 		pvfsp = kmem_zalloc(sizeof (portfop_vfs_t), KM_SLEEP);
1047 		pvfsp->pvfs = vp->v_vfsp;
1048 		list_create(&(pvfsp->pvfs_pvplist), sizeof (portfop_vp_t),
1049 		    offsetof(portfop_vp_t, pvp_pvfsnode));
1050 		pvfsp->pvfs_fsemp = fsemp;
1051 		pvfsp->pvfs_next = pvfsh->pvfshash_pvfsp;
1052 		pvfsh->pvfshash_pvfsp = pvfsp;
1053 	}
1054 
1055 	/*
1056 	 * check if an unmount is in progress.
1057 	 */
1058 	if (!pvfsp->pvfs_unmount) {
1059 		/*
1060 		 * insert the pvp on list.
1061 		 */
1062 		pvp->pvp_pvfsp = pvfsp;
1063 		list_insert_head(&pvfsp->pvfs_pvplist, (void *)pvp);
1064 	} else {
1065 		error = EINVAL;
1066 	}
1067 	mutex_exit(&pvfsh->pvfshash_mutex);
1068 	return (error);
1069 }
1070 
1071 /*
1072  * Installs the portfop_vp_t data structure on the
1073  * vnode. The 'pvp_femp == NULL' indicates it is not
1074  * active. The fem hooks have to be installed.
1075  * The portfop_vp_t is only freed when the vnode gets freed.
1076  */
1077 void
1078 port_install_fopdata(vnode_t *vp)
1079 {
1080 	portfop_vp_t *npvp;
1081 
1082 	npvp = kmem_zalloc(sizeof (*npvp), KM_SLEEP);
1083 	mutex_init(&npvp->pvp_mutex, NULL, MUTEX_DEFAULT, NULL);
1084 	list_create(&npvp->pvp_pfoplist, sizeof (portfop_t),
1085 	    offsetof(portfop_t, pfop_node));
1086 	npvp->pvp_vp = vp;
1087 	/*
1088 	 * If v_fopdata is not null, some other thread beat us to it.
1089 	 */
1090 	if (casptr(&vp->v_fopdata, NULL, npvp) != NULL) {
1091 		mutex_destroy(&npvp->pvp_mutex);
1092 		list_destroy(&npvp->pvp_pfoplist);
1093 		kmem_free(npvp, sizeof (*npvp));
1094 	}
1095 }
1096 
1097 
1098 /*
1099  * Allocate and add a portfop_t to the per port cache. Also add the portfop_t
1100  * to the vnode's list. The association is identified by the object pointer
1101  * address and pid.
1102  */
1103 int
1104 port_pfp_setup(portfop_t **pfpp, port_t *pp, vnode_t *vp, portfop_cache_t *pfcp,
1105 	uintptr_t object, int events, void *user, char *cname, int clen,
1106 	vnode_t *dvp)
1107 {
1108 	portfop_t	*pfp = NULL;
1109 	port_kevent_t	*pkevp;
1110 	fem_t		*femp;
1111 	int		error = 0;
1112 	portfop_vp_t	*pvp;
1113 
1114 
1115 	/*
1116 	 * The port cache mutex is held.
1117 	 */
1118 	*pfpp  = NULL;
1119 
1120 
1121 	/*
1122 	 * At this point the fem monitor is installed.
1123 	 * Allocate a port event structure per vnode association.
1124 	 */
1125 	if (pfp == NULL) {
1126 		if (error = port_alloc_event_local(pp, PORT_SOURCE_FILE,
1127 		    PORT_ALLOC_CACHED, &pkevp)) {
1128 			return (error);
1129 		}
1130 		pfp = kmem_zalloc(sizeof (portfop_t), KM_SLEEP);
1131 		pfp->pfop_pev = pkevp;
1132 	}
1133 
1134 	pfp->pfop_vp = vp;
1135 	pfp->pfop_pid = curproc->p_pid;
1136 	pfp->pfop_pcache = pfcp;
1137 	pfp->pfop_pp = pp;
1138 	pfp->pfop_flags |= PORT_FOP_ACTIVE;
1139 	pfp->pfop_cname = cname;
1140 	pfp->pfop_clen = clen;
1141 	pfp->pfop_dvp = dvp;
1142 	pfp->pfop_object = object;
1143 
1144 	pkevp->portkev_callback = port_fop_callback;
1145 	pkevp->portkev_arg = pfp;
1146 	pkevp->portkev_object = object;
1147 	pkevp->portkev_user = user;
1148 	pkevp->portkev_events = 0;
1149 
1150 	port_pcache_insert(pfcp, pfp);
1151 
1152 	/*
1153 	 * Register a new file events monitor for this file(vnode), if not
1154 	 * done already.
1155 	 */
1156 	if ((pvp = vp->v_fopdata) == NULL) {
1157 		port_install_fopdata(vp);
1158 		pvp = vp->v_fopdata;
1159 	}
1160 
1161 	mutex_enter(&pvp->pvp_mutex);
1162 	/*
1163 	 * if the vnode does not have the file events hooks, install it.
1164 	 */
1165 	if (pvp->pvp_femp == NULL) {
1166 		if ((femp = port_fop_femop()) != NULL) {
1167 			if (!(error = fem_install(pfp->pfop_vp, femp,
1168 			    (void *)vp, OPUNIQ, NULL, NULL))) {
1169 				pvp->pvp_femp = femp;
1170 				/*
1171 				 * add fsem_t hooks to the vfsp and add pvp to
1172 				 * the list of vnodes for this vfs.
1173 				 */
1174 				if (!(error = port_fop_pvfsadd(pvp))) {
1175 					/*
1176 					 * Hold a reference to the vnode since
1177 					 * we successfully installed the hooks.
1178 					 */
1179 					VN_HOLD(vp);
1180 				} else {
1181 					(void) fem_uninstall(vp, femp, vp);
1182 					pvp->pvp_femp = NULL;
1183 				}
1184 			}
1185 		} else {
1186 			error = EINVAL;
1187 		}
1188 	}
1189 
1190 	if (error) {
1191 		/*
1192 		 * pkevp will get freed here.
1193 		 */
1194 		pfp->pfop_cname = NULL;
1195 		port_pcache_remove_fop(pfcp, pfp);
1196 		mutex_exit(&pvp->pvp_mutex);
1197 		return (error);
1198 	}
1199 
1200 	/*
1201 	 * insert the pfp on the vnode's list. After this
1202 	 * events can get delivered.
1203 	 */
1204 	pfp->pfop_events = events;
1205 	port_fop_listinsert_head(pvp, pfp);
1206 
1207 	mutex_exit(&pvp->pvp_mutex);
1208 	/*
1209 	 * Hold the directory vnode since we have a reference now.
1210 	 */
1211 	if (dvp != NULL)
1212 		VN_HOLD(dvp);
1213 	*pfpp = pfp;
1214 	return (0);
1215 }
1216 
1217 vnode_t *
1218 port_resolve_vp(vnode_t *vp)
1219 {
1220 	vnode_t *rvp;
1221 	/*
1222 	 * special case /etc/mnttab(mntfs type). The mntfstype != 0
1223 	 * if mntfs got mounted.
1224 	 */
1225 	if (vfs_mntdummyvp && mntfstype != 0 &&
1226 	    vp->v_vfsp->vfs_fstype == mntfstype) {
1227 		VN_RELE(vp);
1228 		vp = vfs_mntdummyvp;
1229 		VN_HOLD(vfs_mntdummyvp);
1230 	}
1231 
1232 	/*
1233 	 * This should take care of lofs mounted fs systems and nfs4
1234 	 * hardlinks.
1235 	 */
1236 	if ((VOP_REALVP(vp, &rvp, NULL) == 0) && vp != rvp) {
1237 		VN_HOLD(rvp);
1238 		VN_RELE(vp);
1239 		vp = rvp;
1240 	}
1241 	return (vp);
1242 }
1243 
1244 /*
1245  * Register a file events watch on the given file associated to the port *pp.
1246  *
1247  * The association is identified by the object pointer and the pid.
1248  * The events argument contains the events to be monitored for.
1249  *
1250  * The vnode will have a VN_HOLD once the fem hooks are installed.
1251  *
1252  * Every reference(pfp) to the directory vnode will have a VN_HOLD to ensure
1253  * that the directory vnode pointer does not change.
1254  */
1255 int
1256 port_associate_fop(port_t *pp, int source, uintptr_t object, int events,
1257     void *user)
1258 {
1259 	portfop_cache_t	*pfcp;
1260 	vnode_t		*vp, *dvp, *oldvp = NULL, *olddvp = NULL;
1261 	portfop_t	*pfp;
1262 	int		error = 0;
1263 	file_obj_t	fobj;
1264 	void		*objptr;
1265 	char		*cname;
1266 	int		clen;
1267 	int		follow;
1268 
1269 	/*
1270 	 * check that events specified are valid.
1271 	 */
1272 	if ((events & ~FILE_EVENTS_MASK) != 0)
1273 		return (EINVAL);
1274 
1275 	if (get_udatamodel() == DATAMODEL_NATIVE) {
1276 		if (copyin((void *)object, &fobj, sizeof (file_obj_t)))
1277 			return (EFAULT);
1278 		objptr = (void *)&fobj;
1279 #ifdef  _SYSCALL32_IMPL
1280 	} else {
1281 		file_obj32_t	fobj32;
1282 		if (copyin((void *)object, &fobj32, sizeof (file_obj32_t)))
1283 			return (EFAULT);
1284 		objptr = (void *)&fobj32;
1285 #endif  /* _SYSCALL32_IMPL */
1286 	}
1287 
1288 	vp = dvp = NULL;
1289 
1290 	/*
1291 	 * find out if we need to follow symbolic links.
1292 	 */
1293 	follow = !(events & FILE_NOFOLLOW);
1294 	events = events & ~FILE_NOFOLLOW;
1295 
1296 	/*
1297 	 * lookup and find the vnode and its directory vnode of the given
1298 	 * file.
1299 	 */
1300 	if ((error = port_fop_getdvp(objptr, &vp, &dvp, &cname, &clen,
1301 	    follow)) != 0) {
1302 		return (error);
1303 	}
1304 
1305 	if (dvp != NULL) {
1306 		dvp = port_resolve_vp(dvp);
1307 	}
1308 
1309 	/*
1310 	 * Not found
1311 	 */
1312 	if (vp == NULL) {
1313 		error = ENOENT;
1314 		goto errout;
1315 	}
1316 
1317 	vp = port_resolve_vp(vp);
1318 
1319 
1320 	if (vp != NULL && vnevent_support(vp, NULL)) {
1321 		error = ENOTSUP;
1322 		goto errout;
1323 	}
1324 
1325 	/*
1326 	 * Associate this source to the port and get the per port
1327 	 * fop cache pointer. If the source is already associated, it
1328 	 * will just return the cache pointer.
1329 	 */
1330 	if (error = port_fop_associate_source(&pfcp, pp, source)) {
1331 		goto errout;
1332 	}
1333 
1334 	/*
1335 	 * Check if there is an existing association of this file.
1336 	 */
1337 	mutex_enter(&pfcp->pfc_lock);
1338 	pfp = port_cache_lookup_fop(pfcp, curproc->p_pid, object);
1339 
1340 	/*
1341 	 * If it is not the same vnode, just discard it. VN_RELE needs to be
1342 	 * called with no locks held, therefore save vnode pointers and
1343 	 * vn_rele them later.
1344 	 */
1345 	if (pfp != NULL && (pfp->pfop_vp != vp || pfp->pfop_dvp != dvp)) {
1346 		(void) port_remove_fop(pfp, pfcp, 1, NULL, &oldvp, &olddvp);
1347 		pfp = NULL;
1348 	}
1349 
1350 	if (pfp == NULL) {
1351 		vnode_t *tvp, *tdvp;
1352 		portfop_t	*tpfp;
1353 		int error;
1354 
1355 		/*
1356 		 * Add a new association, save the file name and the
1357 		 * directory vnode pointer.
1358 		 */
1359 		if (error = port_pfp_setup(&pfp, pp, vp, pfcp, object,
1360 		    events, user, cname, clen, dvp)) {
1361 			mutex_exit(&pfcp->pfc_lock);
1362 			goto errout;
1363 		}
1364 
1365 		pfp->pfop_callrid = curthread;
1366 		/*
1367 		 * File name used, so make sure we don't free it.
1368 		 */
1369 		cname = NULL;
1370 
1371 		/*
1372 		 * We need to check if the file was removed after the
1373 		 * the lookup and before the fem hooks where added. If
1374 		 * so, return error. The vnode will still exist as we have
1375 		 * a hold on it.
1376 		 *
1377 		 * Drop the cache lock before calling port_fop_getdvp().
1378 		 * port_fop_getdvp() may block either in the vfs layer
1379 		 * or some filesystem.  Therefore there is potential
1380 		 * for deadlock if cache lock is held and if some other
1381 		 * thread is attempting to deliver file events which would
1382 		 * require getting the cache lock, while it may be holding
1383 		 * the filesystem or vfs layer locks.
1384 		 */
1385 		mutex_exit(&pfcp->pfc_lock);
1386 		tvp = NULL;
1387 		if ((error = port_fop_getdvp(objptr, &tvp, NULL,
1388 		    NULL, NULL, follow)) == 0) {
1389 			if (tvp != NULL) {
1390 				tvp = port_resolve_vp(tvp);
1391 				/*
1392 				 * This vnode pointer is just used
1393 				 * for comparison, so rele it
1394 				 */
1395 				VN_RELE(tvp);
1396 			}
1397 		}
1398 
1399 		if (error || tvp == NULL || tvp != vp) {
1400 			/*
1401 			 * Since we dropped the cache lock, make sure
1402 			 * we are still dealing with the same pfp and this
1403 			 * is the thread which registered it.
1404 			 */
1405 			mutex_enter(&pfcp->pfc_lock);
1406 			tpfp = port_cache_lookup_fop(pfcp,
1407 			    curproc->p_pid, object);
1408 
1409 			error = 0;
1410 			if (tpfp == NULL || tpfp != pfp ||
1411 			    pfp->pfop_vp != vp ||
1412 			    pfp->pfop_dvp != dvp ||
1413 			    pfp->pfop_callrid != curthread) {
1414 				/*
1415 				 * Some other event was delivered, the file
1416 				 * watch was removed or reassociated, just
1417 				 * ignore it and leave
1418 				 */
1419 				mutex_exit(&pfcp->pfc_lock);
1420 				goto errout;
1421 			}
1422 
1423 			/*
1424 			 * remove the pfp and fem hooks, if pfp still
1425 			 * active and it is not being removed from
1426 			 * the vnode list. This is checked in
1427 			 * port_remove_fop with the vnode lock held.
1428 			 * The vnode returned is VN_RELE'ed after dropping
1429 			 * the locks.
1430 			 */
1431 			tdvp = tvp = NULL;
1432 			if (port_remove_fop(pfp, pfcp, 0, NULL, &tvp, &tdvp)) {
1433 				/*
1434 				 * The pfp was removed, means no
1435 				 * events where queued. Report the
1436 				 * error now.
1437 				 */
1438 				error = EINVAL;
1439 			}
1440 			mutex_exit(&pfcp->pfc_lock);
1441 			if (tvp != NULL)
1442 				VN_RELE(tvp);
1443 			if (tdvp != NULL)
1444 				VN_RELE(tdvp);
1445 			goto errout;
1446 		}
1447 	} else {
1448 		portfop_vp_t	*pvp = vp->v_fopdata;
1449 
1450 		/*
1451 		 * Re-association of the object.
1452 		 */
1453 		mutex_enter(&pvp->pvp_mutex);
1454 
1455 		/*
1456 		 * remove any queued up event.
1457 		 */
1458 		if (port_remove_done_event(pfp->pfop_pev)) {
1459 			pfp->pfop_flags &= ~PORT_FOP_KEV_ONQ;
1460 		}
1461 
1462 		/*
1463 		 * set new events to watch.
1464 		 */
1465 		pfp->pfop_events = events;
1466 
1467 		/*
1468 		 * If not active, mark it active even if it is being
1469 		 * removed. Then it can send an exception event.
1470 		 *
1471 		 * Move it to the head, as the active ones are only
1472 		 * in the beginning. If removing, the pfp will be on
1473 		 * a temporary list, no need to move it to the front
1474 		 * all the entries will be processed. Some exception
1475 		 * events will be delivered in port_fop_excep();
1476 		 */
1477 		if (!(pfp->pfop_flags & PORT_FOP_ACTIVE)) {
1478 			pfp->pfop_flags |= PORT_FOP_ACTIVE;
1479 			if (!(pfp->pfop_flags & PORT_FOP_REMOVING)) {
1480 				pvp = (portfop_vp_t *)vp->v_fopdata;
1481 				port_fop_listremove(pvp, pfp);
1482 				port_fop_listinsert_head(pvp, pfp);
1483 			}
1484 		}
1485 		pfp->pfop_callrid = curthread;
1486 		mutex_exit(&pvp->pvp_mutex);
1487 		mutex_exit(&pfcp->pfc_lock);
1488 	}
1489 
1490 	/*
1491 	 * Compare time stamps and deliver events.
1492 	 */
1493 	if (vp->v_type != VFIFO) {
1494 		port_check_timestamp(pfcp, vp, dvp, pfp, objptr, object);
1495 	}
1496 
1497 	error = 0;
1498 
1499 	/*
1500 	 *  If we have too many watches on the vnode, discard an
1501 	 *  inactive watch.
1502 	 */
1503 	port_fop_trimpfplist(vp);
1504 
1505 errout:
1506 	/*
1507 	 * Release the hold acquired due to the lookup operation.
1508 	 */
1509 	if (vp != NULL)
1510 		VN_RELE(vp);
1511 	if (dvp != NULL)
1512 		VN_RELE(dvp);
1513 
1514 	if (oldvp != NULL)
1515 		VN_RELE(oldvp);
1516 	if (olddvp != NULL)
1517 		VN_RELE(olddvp);
1518 
1519 	/*
1520 	 * copied file name not used, free it.
1521 	 */
1522 	if (cname != NULL) {
1523 		kmem_free(cname, clen + 1);
1524 	}
1525 	return (error);
1526 }
1527 
1528 
1529 /*
1530  * The port_dissociate_fop() function dissociates the file object
1531  * from the event port and removes any events that are already on the queue.
1532  * Only the owner of the association is allowed to dissociate the file from
1533  * the port. Returns  success (0) if it was found and removed. Otherwise
1534  * ENOENT.
1535  */
1536 int
1537 port_dissociate_fop(port_t *pp, uintptr_t object)
1538 {
1539 	portfop_cache_t	*pfcp;
1540 	portfop_t	*pfp;
1541 	port_source_t	*pse;
1542 	int		active = 0;
1543 	vnode_t		*tvp = NULL, *tdvp = NULL;
1544 
1545 	pse = port_getsrc(pp, PORT_SOURCE_FILE);
1546 
1547 	/*
1548 	 * if this source is not associated or if there is no
1549 	 * cache, nothing to do just return.
1550 	 */
1551 	if (pse == NULL ||
1552 	    (pfcp = (portfop_cache_t *)pse->portsrc_data) == NULL)
1553 		return (EINVAL);
1554 
1555 	/*
1556 	 * Check if this object is on the cache. Only the owner pid
1557 	 * is allowed to dissociate.
1558 	 */
1559 	mutex_enter(&pfcp->pfc_lock);
1560 	pfp = port_cache_lookup_fop(pfcp, curproc->p_pid, object);
1561 	if (pfp == NULL) {
1562 		mutex_exit(&pfcp->pfc_lock);
1563 		return (ENOENT);
1564 	}
1565 
1566 	/*
1567 	 * If this was the last association, it will release
1568 	 * the hold on the vnode. There is a race condition where
1569 	 * the the pfp is being removed due to an exception event
1570 	 * in port_fop_sendevent()->port_fop_excep() and port_remove_fop().
1571 	 * Since port source cache lock is held, port_fop_excep() cannot
1572 	 * complete. The vnode itself will not disappear as long its pfps
1573 	 * have a reference.
1574 	 */
1575 	(void) port_remove_fop(pfp, pfcp, 1, &active, &tvp, &tdvp);
1576 	mutex_exit(&pfcp->pfc_lock);
1577 	if (tvp != NULL)
1578 		VN_RELE(tvp);
1579 	if (tdvp != NULL)
1580 		VN_RELE(tdvp);
1581 	return (active ? 0 : ENOENT);
1582 }
1583 
1584 
1585 /*
1586  * port_close() calls this function to request the PORT_SOURCE_FILE source
1587  * to remove/free all resources allocated and associated with the port.
1588  */
1589 
1590 /* ARGSUSED */
1591 static void
1592 port_close_fop(void *arg, int port, pid_t pid, int lastclose)
1593 {
1594 	port_t		*pp = arg;
1595 	portfop_cache_t	*pfcp;
1596 	portfop_t	**hashtbl;
1597 	portfop_t	*pfp;
1598 	portfop_t	*pfpnext;
1599 	int		index, i;
1600 	port_source_t	*pse;
1601 	vnode_t 	*tdvp = NULL;
1602 	vnode_t		*vpl[PORTFOP_NVP];
1603 
1604 	pse = port_getsrc(pp, PORT_SOURCE_FILE);
1605 
1606 	/*
1607 	 * No source or no cache, nothing to do.
1608 	 */
1609 	if (pse == NULL ||
1610 	    (pfcp = (portfop_cache_t *)pse->portsrc_data) == NULL)
1611 		return;
1612 	/*
1613 	 * Scan the cache and free all allocated portfop_t and port_kevent_t
1614 	 * structures of this pid. Note, no new association for this pid will
1615 	 * be possible as the port is being closed.
1616 	 *
1617 	 * The common case is that the port is not shared and all the entries
1618 	 * are of this pid and have to be freed. Since VN_RELE has to be
1619 	 * called outside the lock, we do it in batches.
1620 	 */
1621 	hashtbl = (portfop_t **)pfcp->pfc_hash;
1622 	index = i = 0;
1623 	bzero(vpl, sizeof (vpl));
1624 	mutex_enter(&pfcp->pfc_lock);
1625 	while (index < PORTFOP_HASHSIZE) {
1626 		pfp = hashtbl[index];
1627 		while (pfp != NULL && i < (PORTFOP_NVP - 1)) {
1628 			pfpnext = pfp->pfop_hashnext;
1629 			if (pid == pfp->pfop_pid) {
1630 				(void) port_remove_fop(pfp, pfcp, 1, NULL,
1631 				    &vpl[i], &tdvp);
1632 				if (vpl[i] != NULL) {
1633 					i++;
1634 				}
1635 				if (tdvp != NULL) {
1636 					vpl[i++] = tdvp;
1637 					tdvp = NULL;
1638 				}
1639 			}
1640 			pfp = pfpnext;
1641 		}
1642 		if (pfp == NULL)
1643 			index++;
1644 		/*
1645 		 * Now call VN_RELE if we have collected enough vnodes or
1646 		 * we have reached the end of the hash table.
1647 		 */
1648 		if (i >= (PORTFOP_NVP - 1) ||
1649 		    (i > 0 && index == PORTFOP_HASHSIZE)) {
1650 			mutex_exit(&pfcp->pfc_lock);
1651 			while (i > 0) {
1652 				VN_RELE(vpl[--i]);
1653 				vpl[i] = NULL;
1654 			}
1655 			mutex_enter(&pfcp->pfc_lock);
1656 		}
1657 	}
1658 
1659 	/*
1660 	 * Due to a race between port_close_fop() and port_fop()
1661 	 * trying to remove the pfp's from the port's cache, it is
1662 	 * possible that some pfp's are still in the process of being
1663 	 * freed so we wait.
1664 	 */
1665 	while (lastclose && pfcp->pfc_objcount) {
1666 		(void) cv_wait_sig(&pfcp->pfc_lclosecv, &pfcp->pfc_lock);
1667 	}
1668 	mutex_exit(&pfcp->pfc_lock);
1669 	/*
1670 	 * last close, free the cache.
1671 	 */
1672 	if (lastclose) {
1673 		ASSERT(pfcp->pfc_objcount == 0);
1674 		pse->portsrc_data = NULL;
1675 		kmem_free(pfcp, sizeof (portfop_cache_t));
1676 	}
1677 }
1678 
1679 /*
1680  * Given the list of associations(watches), it will send exception events,
1681  * if still active, and discard them. The exception events are handled
1682  * separately because, the pfp needs to be removed from the port cache and
1683  * freed as the vnode's identity is changing or being removed. To remove
1684  * the pfp from the port's cache, we need to hold the cache lock (pfc_lock).
1685  * The lock order is pfc_lock -> pvp_mutex(vnode's) mutex and that is why
1686  * the cache's lock cannot be acquired in port_fop_sendevent().
1687  */
1688 static void
1689 port_fop_excep(list_t *tlist, int op)
1690 {
1691 	portfop_t	*pfp;
1692 	portfop_cache_t *pfcp;
1693 	port_t	*pp;
1694 	port_kevent_t	*pkevp;
1695 	vnode_t		*tdvp;
1696 	int		error = 0;
1697 
1698 	while (pfp = (portfop_t *)list_head(tlist)) {
1699 		int removed = 0;
1700 		/*
1701 		 * remove from the temp list. Since PORT_FOP_REMOVING is
1702 		 * set, no other thread should attempt to perform a
1703 		 * list_remove on this pfp.
1704 		 */
1705 		list_remove(tlist, pfp);
1706 
1707 		pfcp = pfp->pfop_pcache;
1708 		mutex_enter(&pfcp->pfc_lock);
1709 
1710 		/*
1711 		 * Remove the event from the port queue if it was queued up.
1712 		 * No need to clear the PORT_FOP_KEV_ONQ flag as this pfp is
1713 		 * no longer on the vnode's list.
1714 		 */
1715 		if ((pfp->pfop_flags & PORT_FOP_KEV_ONQ)) {
1716 			removed = port_remove_done_event(pfp->pfop_pev);
1717 		}
1718 
1719 		/*
1720 		 * If still active or the event was queued up and
1721 		 * had not been collected yet, send an EXCEPTION event.
1722 		 */
1723 		if (pfp->pfop_flags & (PORT_FOP_ACTIVE) || removed) {
1724 			pp = pfp->pfop_pp;
1725 			/*
1726 			 * Allocate a port_kevent_t non cached to send this
1727 			 * event since we will be de-registering.
1728 			 * The port_kevent_t cannot be pointing back to the
1729 			 * pfp anymore.
1730 			 */
1731 			pfp->pfop_flags &= ~PORT_FOP_ACTIVE;
1732 			error = port_alloc_event_local(pp, PORT_SOURCE_FILE,
1733 			    PORT_ALLOC_DEFAULT, &pkevp);
1734 			if (!error) {
1735 
1736 				pkevp->portkev_callback = port_fop_callback;
1737 				pkevp->portkev_arg = NULL;
1738 				pkevp->portkev_object =
1739 				    pfp->pfop_pev->portkev_object;
1740 				pkevp->portkev_user =
1741 				    pfp->pfop_pev->portkev_user;
1742 				/*
1743 				 * Copy the pid of the watching process.
1744 				 */
1745 				pkevp->portkev_pid =
1746 				    pfp->pfop_pev->portkev_pid;
1747 				pkevp->portkev_events = op;
1748 				port_send_event(pkevp);
1749 			}
1750 		}
1751 		/*
1752 		 * At this point the pfp has been removed from the vnode's
1753 		 * list its cached port_kevent_t is not on the done queue.
1754 		 * Remove the pfp and free it from the cache.
1755 		 */
1756 		tdvp = pfp->pfop_dvp;
1757 		port_pcache_remove_fop(pfcp, pfp);
1758 		mutex_exit(&pfcp->pfc_lock);
1759 		if (tdvp != NULL)
1760 			VN_RELE(tdvp);
1761 	}
1762 }
1763 
1764 /*
1765  * Send the file events to all of the processes watching this
1766  * vnode. In case of hard links, the directory vnode pointer and
1767  * the file name are compared. If the names match, then the specified
1768  * event is sent or else, the FILE_ATTRIB event is sent, This is the
1769  * documented behavior.
1770  */
1771 void
1772 port_fop_sendevent(vnode_t *vp, int events, vnode_t *dvp, char *cname)
1773 {
1774 	port_kevent_t	*pkevp;
1775 	portfop_t	*pfp, *npfp;
1776 	portfop_vp_t	*pvp;
1777 	list_t		tmplist;
1778 	int		removeall = 0;
1779 
1780 	pvp = (portfop_vp_t *)vp->v_fopdata;
1781 	mutex_enter(&pvp->pvp_mutex);
1782 
1783 	/*
1784 	 * Check if the list is empty.
1785 	 *
1786 	 * All entries have been removed by some other thread.
1787 	 * The vnode may be still active and we got called,
1788 	 * but some other thread is in the process of removing the hooks.
1789 	 */
1790 	if (!list_head(&pvp->pvp_pfoplist)) {
1791 		mutex_exit(&pvp->pvp_mutex);
1792 		return;
1793 	}
1794 
1795 	if ((events & (FILE_EXCEPTION))) {
1796 		/*
1797 		 * If it is an event for which we are going to remove
1798 		 * the watches so just move it a temporary list and
1799 		 * release this vnode.
1800 		 */
1801 		list_create(&tmplist, sizeof (portfop_t),
1802 		    offsetof(portfop_t, pfop_node));
1803 
1804 		/*
1805 		 * If it is an UNMOUNT, MOUNTEDOVER or no file name has been
1806 		 * passed for an exception event, all associations need to be
1807 		 * removed.
1808 		 */
1809 		if (dvp == NULL || cname == NULL) {
1810 			removeall = 1;
1811 		}
1812 	}
1813 
1814 	if (!removeall) {
1815 		/*
1816 		 * All the active ones are in the beginning of the list.
1817 		 */
1818 		for (pfp = (portfop_t *)list_head(&pvp->pvp_pfoplist);
1819 		    pfp && pfp->pfop_flags & PORT_FOP_ACTIVE; pfp = npfp) {
1820 			int levents = events;
1821 
1822 			npfp = list_next(&pvp->pvp_pfoplist, pfp);
1823 			/*
1824 			 * Hard links case - If the file is being
1825 			 * removed/renamed, and the name matches
1826 			 * the watched file, then it is an EXCEPTION
1827 			 * event or else it will be just a FILE_ATTRIB.
1828 			 */
1829 			if ((events & (FILE_EXCEPTION))) {
1830 				ASSERT(dvp != NULL && cname != NULL);
1831 				if (pfp->pfop_dvp == NULL ||
1832 				    (pfp->pfop_dvp == dvp &&
1833 				    (strcmp(cname, pfp->pfop_cname) == 0))) {
1834 					/*
1835 					 * It is an exception event, move it
1836 					 * to temp list and process it later.
1837 					 * Note we don't set the pfp->pfop_vp
1838 					 * to NULL even thought it has been
1839 					 * removed from the vnode's list. This
1840 					 * pointer is referenced in
1841 					 * port_remove_fop(). The vnode it
1842 					 * self cannot disappear until this
1843 					 * pfp gets removed and freed.
1844 					 */
1845 					port_fop_listremove(pvp, pfp);
1846 					list_insert_tail(&tmplist, (void *)pfp);
1847 					pfp->pfop_flags  |= PORT_FOP_REMOVING;
1848 					continue;
1849 				} else {
1850 					levents = FILE_ATTRIB;
1851 				}
1852 
1853 			}
1854 
1855 			if (pfp->pfop_events & levents) {
1856 				/*
1857 				 * deactivate and move it to the tail.
1858 				 * If the pfp was active, it cannot be
1859 				 * on the port's done queue.
1860 				 */
1861 				pfp->pfop_flags &= ~PORT_FOP_ACTIVE;
1862 				port_fop_listremove(pvp, pfp);
1863 				port_fop_listinsert_tail(pvp, pfp);
1864 
1865 				pkevp = pfp->pfop_pev;
1866 				pkevp->portkev_events |=
1867 				    (levents & pfp->pfop_events);
1868 				port_send_event(pkevp);
1869 				pfp->pfop_flags |= PORT_FOP_KEV_ONQ;
1870 			}
1871 		}
1872 	}
1873 
1874 
1875 	if ((events & (FILE_EXCEPTION))) {
1876 		if (!removeall) {
1877 			/*
1878 			 * Check the inactive associations and remove them if
1879 			 * the file name matches.
1880 			 */
1881 			for (; pfp; pfp = npfp) {
1882 				npfp = list_next(&pvp->pvp_pfoplist, pfp);
1883 				if (dvp == NULL || cname == NULL ||
1884 				    pfp->pfop_dvp == NULL ||
1885 				    (pfp->pfop_dvp == dvp &&
1886 				    (strcmp(cname, pfp->pfop_cname) == 0))) {
1887 					port_fop_listremove(pvp, pfp);
1888 					list_insert_tail(&tmplist, (void *)pfp);
1889 					pfp->pfop_flags  |= PORT_FOP_REMOVING;
1890 				}
1891 			}
1892 		} else {
1893 			/*
1894 			 * Can be optimized to avoid two pass over this list
1895 			 * by having a flag in the vnode's portfop_vp_t
1896 			 * structure to indicate that it is going away,
1897 			 * Or keep the list short by reusing inactive watches.
1898 			 */
1899 			port_fop_listmove(pvp, &tmplist);
1900 			for (pfp = (portfop_t *)list_head(&tmplist);
1901 			    pfp; pfp = list_next(&tmplist, pfp)) {
1902 				pfp->pfop_flags |= PORT_FOP_REMOVING;
1903 			}
1904 		}
1905 
1906 		/*
1907 		 * Uninstall the fem hooks if there are no more associations.
1908 		 * This will release the pvp mutex.
1909 		 *
1910 		 * Even thought all entries may have been removed,
1911 		 * the vnode itself cannot disappear as there will be a
1912 		 * hold on it due to this call to port_fop_sendevent. This is
1913 		 * important to syncronize with a port_dissociate_fop() call
1914 		 * that may be attempting to remove an object from the vnode's.
1915 		 */
1916 		if (port_fop_femuninstall(vp))
1917 			VN_RELE(vp);
1918 
1919 		/*
1920 		 * Send exception events and discard the watch entries.
1921 		 */
1922 		port_fop_excep(&tmplist, events);
1923 		list_destroy(&tmplist);
1924 
1925 	} else {
1926 		mutex_exit(&pvp->pvp_mutex);
1927 
1928 		/*
1929 		 * trim the list.
1930 		 */
1931 		port_fop_trimpfplist(vp);
1932 	}
1933 }
1934 
1935 /*
1936  * Given the file operation, map it to the event types and send.
1937  */
1938 void
1939 port_fop(vnode_t *vp, int op, int retval)
1940 {
1941 	int event = 0;
1942 	/*
1943 	 * deliver events only if the operation was successful.
1944 	 */
1945 	if (retval)
1946 		return;
1947 
1948 	/*
1949 	 * These events occurring on the watched file.
1950 	 */
1951 	if (op & FOP_MODIFIED_MASK) {
1952 		event  = FILE_MODIFIED;
1953 	}
1954 	if (op & FOP_ACCESS_MASK) {
1955 		event  |= FILE_ACCESS;
1956 	}
1957 	if (op & FOP_ATTRIB_MASK) {
1958 		event  |= FILE_ATTRIB;
1959 	}
1960 
1961 	if (event) {
1962 		port_fop_sendevent(vp, 	event, NULL, NULL);
1963 	}
1964 }
1965 
1966 static int port_forceunmount(vfs_t *vfsp)
1967 {
1968 	char *fsname = vfssw[vfsp->vfs_fstype].vsw_name;
1969 
1970 	if (fsname == NULL) {
1971 		return (0);
1972 	}
1973 
1974 	if (strcmp(fsname, MNTTYPE_NFS) == 0) {
1975 		return (1);
1976 	}
1977 
1978 	if (strcmp(fsname, MNTTYPE_NFS3) == 0) {
1979 		return (1);
1980 	}
1981 
1982 	if (strcmp(fsname, MNTTYPE_NFS4) == 0) {
1983 		return (1);
1984 	}
1985 	return (0);
1986 }
1987 /*
1988  * ----- the unmount filesystem op(fsem) hook.
1989  */
1990 int
1991 port_fop_unmount(fsemarg_t *vf, int flag, cred_t *cr)
1992 {
1993 	vfs_t	*vfsp = (vfs_t *)vf->fa_fnode->fn_available;
1994 	kmutex_t	*mtx;
1995 	portfop_vfs_t	*pvfsp, **ppvfsp;
1996 	portfop_vp_t	*pvp;
1997 	int error;
1998 	int fmfs;
1999 
2000 	fmfs = port_forceunmount(vfsp);
2001 
2002 	mtx = &(portvfs_hash[PORTFOP_PVFSHASH(vfsp)].pvfshash_mutex);
2003 	ppvfsp = &(portvfs_hash[PORTFOP_PVFSHASH(vfsp)].pvfshash_pvfsp);
2004 	pvfsp = NULL;
2005 	mutex_enter(mtx);
2006 	/*
2007 	 * since this fsem hook is triggered, the vfsp has to be on
2008 	 * the hash list.
2009 	 */
2010 	for (pvfsp = *ppvfsp; pvfsp->pvfs != vfsp; pvfsp = pvfsp->pvfs_next)
2011 	;
2012 
2013 	/*
2014 	 * For some of the filesystems, allow unmounts to proceed only if
2015 	 * there are no files being watched or it is a forced unmount.
2016 	 */
2017 	if (fmfs && !(flag & MS_FORCE) &&
2018 	    !list_is_empty(&pvfsp->pvfs_pvplist)) {
2019 		mutex_exit(mtx);
2020 		return (EBUSY);
2021 	}
2022 
2023 	/*
2024 	 * Indicate that the unmount is in process. Don't remove it yet.
2025 	 * The underlying filesystem unmount routine sets the VFS_UNMOUNTED
2026 	 * flag on the vfs_t structure. But we call the filesystem unmount
2027 	 * routine after removing all the file watches for this filesystem,
2028 	 * otherwise the unmount will fail due to active vnodes.
2029 	 * Meanwhile setting pvfsp->unmount = 1 will prevent any thread
2030 	 * attempting to add a file watch.
2031 	 */
2032 	pvfsp->pvfs_unmount = 1;
2033 	mutex_exit(mtx);
2034 
2035 	/*
2036 	 * uninstall the fsem hooks.
2037 	 */
2038 	(void) fsem_uninstall(vfsp, (fsem_t *)pvfsp->pvfs_fsemp, vfsp);
2039 
2040 	while (pvp = list_head(&pvfsp->pvfs_pvplist)) {
2041 		list_remove(&pvfsp->pvfs_pvplist, pvp);
2042 		/*
2043 		 * This should send an UNMOUNTED event to all the
2044 		 * watched vnode of this filesystem and uninstall
2045 		 * the fem hooks. We release the hold on the vnode here
2046 		 * because port_fop_femuninstall() will not do it if
2047 		 * unmount is in process.
2048 		 */
2049 		port_fop_sendevent(pvp->pvp_vp, UNMOUNTED, NULL, NULL);
2050 		VN_RELE(pvp->pvp_vp);
2051 	}
2052 
2053 	error = vfsnext_unmount(vf, flag, cr);
2054 
2055 	/*
2056 	 * we free the pvfsp after the unmount has been completed.
2057 	 */
2058 	mutex_enter(mtx);
2059 	for (; *ppvfsp && (*ppvfsp)->pvfs != vfsp;
2060 	    ppvfsp = &(*ppvfsp)->pvfs_next)
2061 	;
2062 
2063 	/*
2064 	 * remove and free it.
2065 	 */
2066 	ASSERT(list_head(&pvfsp->pvfs_pvplist) == NULL);
2067 	if (*ppvfsp) {
2068 		pvfsp = *ppvfsp;
2069 		*ppvfsp = pvfsp->pvfs_next;
2070 	}
2071 	mutex_exit(mtx);
2072 	kmem_free(pvfsp, sizeof (portfop_vfs_t));
2073 	return (error);
2074 }
2075 
2076 /*
2077  * ------------------------------file op hooks--------------------------
2078  * The O_TRUNC operation is caught with the VOP_SETATTR(AT_SIZE) call.
2079  */
2080 static int
2081 port_fop_open(femarg_t *vf, int mode, cred_t *cr, caller_context_t *ct)
2082 {
2083 	int		retval;
2084 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2085 
2086 	retval = vnext_open(vf, mode, cr, ct);
2087 	port_fop(vp, FOP_FILE_OPEN, retval);
2088 	return (retval);
2089 }
2090 
2091 static int
2092 port_fop_write(femarg_t *vf, struct uio *uiop, int ioflag, struct cred *cr,
2093     caller_context_t *ct)
2094 {
2095 	int		retval;
2096 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2097 
2098 	retval =  vnext_write(vf, uiop, ioflag, cr, ct);
2099 	port_fop(vp, FOP_FILE_WRITE, retval);
2100 	return (retval);
2101 }
2102 
2103 static int
2104 port_fop_map(femarg_t *vf, offset_t off, struct as *as, caddr_t *addrp,
2105     size_t len, uchar_t prot, uchar_t maxport, uint_t flags, cred_t *cr,
2106     caller_context_t *ct)
2107 {
2108 	int		retval;
2109 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2110 
2111 	retval =  vnext_map(vf, off, as, addrp, len, prot, maxport,
2112 	    flags, cr, ct);
2113 	port_fop(vp, FOP_FILE_MAP, retval);
2114 	return (retval);
2115 }
2116 
2117 static int
2118 port_fop_read(femarg_t *vf, struct uio *uiop, int ioflag, struct cred *cr,
2119     caller_context_t *ct)
2120 {
2121 	int		retval;
2122 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2123 
2124 	retval =  vnext_read(vf, uiop, ioflag, cr, ct);
2125 	port_fop(vp, FOP_FILE_READ, retval);
2126 	return (retval);
2127 }
2128 
2129 
2130 /*
2131  * AT_SIZE - is for the open(O_TRUNC) case.
2132  */
2133 int
2134 port_fop_setattr(femarg_t *vf, vattr_t *vap, int flags, cred_t *cr,
2135     caller_context_t *ct)
2136 {
2137 	int		retval;
2138 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2139 	int		events = 0;
2140 
2141 	retval = vnext_setattr(vf, vap, flags, cr, ct);
2142 	if (vap->va_mask & (AT_SIZE|AT_MTIME)) {
2143 		events |= FOP_FILE_SETATTR_MTIME;
2144 	}
2145 	if (vap->va_mask & AT_ATIME) {
2146 		events |= FOP_FILE_SETATTR_ATIME;
2147 	}
2148 	events |= FOP_FILE_SETATTR_CTIME;
2149 
2150 	port_fop(vp, events, retval);
2151 	return (retval);
2152 }
2153 
2154 int
2155 port_fop_create(femarg_t *vf, char *name, vattr_t *vap, vcexcl_t excl,
2156     int mode, vnode_t **vpp, cred_t *cr, int flag,
2157     caller_context_t *ct, vsecattr_t *vsecp)
2158 {
2159 	int		retval, got = 1;
2160 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2161 	vattr_t		vatt, vatt1;
2162 
2163 	/*
2164 	 * If the file already exists, then there will be no change
2165 	 * to the directory. Therefore, we need to compare the
2166 	 * modification time of the directory to determine if the
2167 	 * file was actually created.
2168 	 */
2169 	vatt.va_mask = AT_ATIME|AT_MTIME|AT_CTIME;
2170 	if (VOP_GETATTR(vp, &vatt, 0, CRED(), ct)) {
2171 		got = 0;
2172 	}
2173 	retval = vnext_create(vf, name, vap, excl, mode, vpp, cr,
2174 	    flag, ct, vsecp);
2175 
2176 	vatt1.va_mask = AT_ATIME|AT_MTIME|AT_CTIME;
2177 	if (got && !VOP_GETATTR(vp, &vatt1, 0, CRED(), ct)) {
2178 		if ((vatt1.va_mtime.tv_sec > vatt.va_mtime.tv_sec ||
2179 		    (vatt1.va_mtime.tv_sec = vatt.va_mtime.tv_sec &&
2180 		    vatt1.va_mtime.tv_nsec > vatt.va_mtime.tv_nsec))) {
2181 			/*
2182 			 * File was created.
2183 			 */
2184 			port_fop(vp, FOP_FILE_CREATE, retval);
2185 		}
2186 	}
2187 	return (retval);
2188 }
2189 
2190 int
2191 port_fop_remove(femarg_t *vf, char *nm, cred_t *cr, caller_context_t *ct,
2192     int flags)
2193 {
2194 	int		retval;
2195 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2196 
2197 	retval = vnext_remove(vf, nm, cr, ct, flags);
2198 	port_fop(vp, FOP_FILE_REMOVE, retval);
2199 	return (retval);
2200 }
2201 
2202 int
2203 port_fop_link(femarg_t *vf, vnode_t *svp, char *tnm, cred_t *cr,
2204     caller_context_t *ct, int flags)
2205 {
2206 	int		retval;
2207 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2208 
2209 	retval = vnext_link(vf, svp, tnm, cr, ct, flags);
2210 	port_fop(vp, FOP_FILE_LINK, retval);
2211 	return (retval);
2212 }
2213 
2214 /*
2215  * Rename operation is allowed only when from and to directories are
2216  * on the same filesystem. This is checked in vn_rename().
2217  * The target directory is notified thru a VNEVENT by the filesystem
2218  * if the source dir != target dir.
2219  */
2220 int
2221 port_fop_rename(femarg_t *vf, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr,
2222     caller_context_t *ct, int flags)
2223 {
2224 	int		retval;
2225 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2226 
2227 	retval = vnext_rename(vf, snm, tdvp, tnm, cr, ct, flags);
2228 	port_fop(vp, FOP_FILE_RENAMESRC, retval);
2229 	return (retval);
2230 }
2231 
2232 int
2233 port_fop_mkdir(femarg_t *vf, char *dirname, vattr_t *vap, vnode_t **vpp,
2234     cred_t *cr, caller_context_t *ct, int flags, vsecattr_t *vsecp)
2235 {
2236 	int		retval;
2237 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2238 
2239 	retval = vnext_mkdir(vf, dirname, vap, vpp, cr, ct, flags, vsecp);
2240 	port_fop(vp, FOP_FILE_MKDIR, retval);
2241 	return (retval);
2242 }
2243 
2244 int
2245 port_fop_rmdir(femarg_t *vf, char *nm, vnode_t *cdir, cred_t *cr,
2246     caller_context_t *ct, int flags)
2247 {
2248 	int		retval;
2249 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2250 
2251 	retval = vnext_rmdir(vf, nm, cdir, cr, ct, flags);
2252 	port_fop(vp, FOP_FILE_RMDIR, retval);
2253 	return (retval);
2254 }
2255 
2256 int
2257 port_fop_readdir(femarg_t *vf, uio_t *uiop, cred_t *cr, int *eofp,
2258     caller_context_t *ct, int flags)
2259 {
2260 	int		retval;
2261 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2262 
2263 	retval = vnext_readdir(vf, uiop, cr, eofp, ct, flags);
2264 	port_fop(vp, FOP_FILE_READDIR, retval);
2265 	return (retval);
2266 }
2267 
2268 int
2269 port_fop_symlink(femarg_t *vf, char *linkname, vattr_t *vap, char *target,
2270     cred_t *cr, caller_context_t *ct, int flags)
2271 {
2272 	int		retval;
2273 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2274 
2275 	retval = vnext_symlink(vf, linkname, vap, target, cr, ct, flags);
2276 	port_fop(vp, FOP_FILE_SYMLINK, retval);
2277 	return (retval);
2278 }
2279 
2280 /*
2281  * acl, facl call this.
2282  */
2283 int
2284 port_fop_setsecattr(femarg_t *vf, vsecattr_t *vsap, int flags, cred_t *cr,
2285     caller_context_t *ct)
2286 {
2287 	int	retval;
2288 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2289 	retval = vnext_setsecattr(vf, vsap, flags, cr, ct);
2290 	port_fop(vp, FOP_FILE_SETSECATTR, retval);
2291 	return (retval);
2292 }
2293 
2294 /*
2295  * these are events on the watched file/directory
2296  */
2297 int
2298 port_fop_vnevent(femarg_t *vf, vnevent_t vnevent, vnode_t *dvp, char *name,
2299     caller_context_t *ct)
2300 {
2301 	vnode_t		*vp = (vnode_t *)vf->fa_fnode->fn_available;
2302 
2303 	switch (vnevent) {
2304 	case	VE_RENAME_SRC:
2305 			port_fop_sendevent(vp, FILE_RENAME_FROM, dvp, name);
2306 		break;
2307 	case	VE_RENAME_DEST:
2308 			port_fop_sendevent(vp, FILE_RENAME_TO, dvp, name);
2309 		break;
2310 	case	VE_REMOVE:
2311 			port_fop_sendevent(vp, FILE_DELETE, dvp, name);
2312 		break;
2313 	case	VE_RMDIR:
2314 			port_fop_sendevent(vp, FILE_DELETE, dvp, name);
2315 		break;
2316 	case	VE_CREATE:
2317 			port_fop_sendevent(vp, FILE_MODIFIED|FILE_ATTRIB,
2318 			    NULL, NULL);
2319 		break;
2320 	case	VE_LINK:
2321 			port_fop_sendevent(vp, FILE_ATTRIB, NULL, NULL);
2322 		break;
2323 
2324 	case	VE_RENAME_DEST_DIR:
2325 			port_fop_sendevent(vp, FILE_MODIFIED|FILE_ATTRIB,
2326 			    NULL, NULL);
2327 		break;
2328 
2329 	case	VE_MOUNTEDOVER:
2330 			port_fop_sendevent(vp, MOUNTEDOVER, NULL, NULL);
2331 		break;
2332 	default:
2333 		break;
2334 	}
2335 	return (vnext_vnevent(vf, vnevent, dvp, name, ct));
2336 }
2337