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 (c) 1986, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2017 by Delphix. All rights reserved.
24  */
25 
26 /*
27  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
28  *	All Rights Reserved
29  */
30 
31 #include <sys/param.h>
32 #include <sys/types.h>
33 #include <sys/systm.h>
34 #include <sys/thread.h>
35 #include <sys/t_lock.h>
36 #include <sys/time.h>
37 #include <sys/vnode.h>
38 #include <sys/vfs.h>
39 #include <sys/errno.h>
40 #include <sys/buf.h>
41 #include <sys/stat.h>
42 #include <sys/cred.h>
43 #include <sys/kmem.h>
44 #include <sys/debug.h>
45 #include <sys/dnlc.h>
46 #include <sys/vmsystm.h>
47 #include <sys/flock.h>
48 #include <sys/share.h>
49 #include <sys/cmn_err.h>
50 #include <sys/tiuser.h>
51 #include <sys/sysmacros.h>
52 #include <sys/callb.h>
53 #include <sys/acl.h>
54 #include <sys/kstat.h>
55 #include <sys/signal.h>
56 #include <sys/disp.h>
57 #include <sys/atomic.h>
58 #include <sys/list.h>
59 #include <sys/sdt.h>
60 
61 #include <rpc/types.h>
62 #include <rpc/xdr.h>
63 #include <rpc/auth.h>
64 #include <rpc/clnt.h>
65 
66 #include <nfs/nfs.h>
67 #include <nfs/nfs_clnt.h>
68 #include <nfs/nfs_acl.h>
69 
70 #include <nfs/nfs4.h>
71 #include <nfs/rnode4.h>
72 #include <nfs/nfs4_clnt.h>
73 
74 #include <vm/hat.h>
75 #include <vm/as.h>
76 #include <vm/page.h>
77 #include <vm/pvn.h>
78 #include <vm/seg.h>
79 #include <vm/seg_map.h>
80 #include <vm/seg_vn.h>
81 
82 #include <sys/ddi.h>
83 
84 /*
85  * Arguments to page-flush thread.
86  */
87 typedef struct {
88 	vnode_t *vp;
89 	cred_t *cr;
90 } pgflush_t;
91 
92 #ifdef DEBUG
93 int nfs4_client_lease_debug;
94 int nfs4_sharedfh_debug;
95 int nfs4_fname_debug;
96 
97 /* temporary: panic if v_type is inconsistent with r_attr va_type */
98 int nfs4_vtype_debug;
99 
100 uint_t nfs4_tsd_key;
101 #endif
102 
103 static time_t	nfs4_client_resumed = 0;
104 static	callb_id_t cid = 0;
105 
106 static int	nfs4renew(nfs4_server_t *);
107 static void	nfs4_attrcache_va(vnode_t *, nfs4_ga_res_t *, int);
108 static void	nfs4_pgflush_thread(pgflush_t *);
109 
110 static boolean_t nfs4_client_cpr_callb(void *, int);
111 
112 struct mi4_globals {
113 	kmutex_t	mig_lock;  /* lock protecting mig_list */
114 	list_t		mig_list;  /* list of NFS v4 mounts in zone */
115 	boolean_t	mig_destructor_called;
116 };
117 
118 static zone_key_t mi4_list_key;
119 
120 /*
121  * Attributes caching:
122  *
123  * Attributes are cached in the rnode in struct vattr form.
124  * There is a time associated with the cached attributes (r_time_attr_inval)
125  * which tells whether the attributes are valid. The time is initialized
126  * to the difference between current time and the modify time of the vnode
127  * when new attributes are cached. This allows the attributes for
128  * files that have changed recently to be timed out sooner than for files
129  * that have not changed for a long time. There are minimum and maximum
130  * timeout values that can be set per mount point.
131  */
132 
133 /*
134  * If a cache purge is in progress, wait for it to finish.
135  *
136  * The current thread must not be in the middle of an
137  * nfs4_start_op/nfs4_end_op region.  Otherwise, there could be a deadlock
138  * between this thread, a recovery thread, and the page flush thread.
139  */
140 int
nfs4_waitfor_purge_complete(vnode_t * vp)141 nfs4_waitfor_purge_complete(vnode_t *vp)
142 {
143 	rnode4_t *rp;
144 	k_sigset_t smask;
145 
146 	rp = VTOR4(vp);
147 	if ((rp->r_serial != NULL && rp->r_serial != curthread) ||
148 	    ((rp->r_flags & R4PGFLUSH) && rp->r_pgflush != curthread)) {
149 		mutex_enter(&rp->r_statelock);
150 		sigintr(&smask, VTOMI4(vp)->mi_flags & MI4_INT);
151 		while ((rp->r_serial != NULL && rp->r_serial != curthread) ||
152 		    ((rp->r_flags & R4PGFLUSH) &&
153 		    rp->r_pgflush != curthread)) {
154 			if (!cv_wait_sig(&rp->r_cv, &rp->r_statelock)) {
155 				sigunintr(&smask);
156 				mutex_exit(&rp->r_statelock);
157 				return (EINTR);
158 			}
159 		}
160 		sigunintr(&smask);
161 		mutex_exit(&rp->r_statelock);
162 	}
163 	return (0);
164 }
165 
166 /*
167  * Validate caches by checking cached attributes. If they have timed out,
168  * then get new attributes from the server.  As a side effect, cache
169  * invalidation is done if the attributes have changed.
170  *
171  * If the attributes have not timed out and if there is a cache
172  * invalidation being done by some other thread, then wait until that
173  * thread has completed the cache invalidation.
174  */
175 int
nfs4_validate_caches(vnode_t * vp,cred_t * cr)176 nfs4_validate_caches(vnode_t *vp, cred_t *cr)
177 {
178 	int error;
179 	nfs4_ga_res_t gar;
180 
181 	if (ATTRCACHE4_VALID(vp)) {
182 		error = nfs4_waitfor_purge_complete(vp);
183 		if (error)
184 			return (error);
185 		return (0);
186 	}
187 
188 	return (nfs4_getattr_otw(vp, &gar, cr, 0));
189 }
190 
191 /*
192  * Fill in attribute from the cache.
193  * If valid, then return 0 to indicate that no error occurred,
194  * otherwise return 1 to indicate that an error occurred.
195  */
196 static int
nfs4_getattr_cache(vnode_t * vp,struct vattr * vap)197 nfs4_getattr_cache(vnode_t *vp, struct vattr *vap)
198 {
199 	rnode4_t *rp;
200 
201 	rp = VTOR4(vp);
202 	mutex_enter(&rp->r_statelock);
203 	mutex_enter(&rp->r_statev4_lock);
204 	if (ATTRCACHE4_VALID(vp)) {
205 		mutex_exit(&rp->r_statev4_lock);
206 		/*
207 		 * Cached attributes are valid
208 		 */
209 		*vap = rp->r_attr;
210 		mutex_exit(&rp->r_statelock);
211 		return (0);
212 	}
213 	mutex_exit(&rp->r_statev4_lock);
214 	mutex_exit(&rp->r_statelock);
215 	return (1);
216 }
217 
218 
219 /*
220  * If returned error is ESTALE flush all caches.  The nfs4_purge_caches()
221  * call is synchronous because all the pages were invalidated by the
222  * nfs4_invalidate_pages() call.
223  */
224 void
nfs4_purge_stale_fh(int errno,vnode_t * vp,cred_t * cr)225 nfs4_purge_stale_fh(int errno, vnode_t *vp, cred_t *cr)
226 {
227 	struct rnode4 *rp = VTOR4(vp);
228 
229 	/* Ensure that the ..._end_op() call has been done */
230 	ASSERT(tsd_get(nfs4_tsd_key) == NULL);
231 
232 	if (errno != ESTALE)
233 		return;
234 
235 	mutex_enter(&rp->r_statelock);
236 	rp->r_flags |= R4STALE;
237 	if (!rp->r_error)
238 		rp->r_error = errno;
239 	mutex_exit(&rp->r_statelock);
240 	if (nfs4_has_pages(vp))
241 		nfs4_invalidate_pages(vp, (u_offset_t)0, cr);
242 	nfs4_purge_caches(vp, NFS4_PURGE_DNLC, cr, FALSE);
243 }
244 
245 /*
246  * Purge all of the various NFS `data' caches.  If "asyncpg" is TRUE, the
247  * page purge is done asynchronously.
248  */
249 void
nfs4_purge_caches(vnode_t * vp,int purge_dnlc,cred_t * cr,int asyncpg)250 nfs4_purge_caches(vnode_t *vp, int purge_dnlc, cred_t *cr, int asyncpg)
251 {
252 	rnode4_t *rp;
253 	char *contents;
254 	vnode_t *xattr;
255 	int size;
256 	int pgflush;			/* are we the page flush thread? */
257 
258 	/*
259 	 * Purge the DNLC for any entries which refer to this file.
260 	 */
261 	if (vp->v_count > 1 &&
262 	    (vp->v_type == VDIR || purge_dnlc == NFS4_PURGE_DNLC))
263 		dnlc_purge_vp(vp);
264 
265 	/*
266 	 * Clear any readdir state bits and purge the readlink response cache.
267 	 */
268 	rp = VTOR4(vp);
269 	mutex_enter(&rp->r_statelock);
270 	rp->r_flags &= ~R4LOOKUP;
271 	contents = rp->r_symlink.contents;
272 	size = rp->r_symlink.size;
273 	rp->r_symlink.contents = NULL;
274 
275 	xattr = rp->r_xattr_dir;
276 	rp->r_xattr_dir = NULL;
277 
278 	/*
279 	 * Purge pathconf cache too.
280 	 */
281 	rp->r_pathconf.pc4_xattr_valid = 0;
282 	rp->r_pathconf.pc4_cache_valid = 0;
283 
284 	pgflush = (curthread == rp->r_pgflush);
285 	mutex_exit(&rp->r_statelock);
286 
287 	if (contents != NULL) {
288 
289 		kmem_free((void *)contents, size);
290 	}
291 
292 	if (xattr != NULL)
293 		VN_RELE(xattr);
294 
295 	/*
296 	 * Flush the page cache.  If the current thread is the page flush
297 	 * thread, don't initiate a new page flush.  There's no need for
298 	 * it, and doing it correctly is hard.
299 	 */
300 	if (nfs4_has_pages(vp) && !pgflush) {
301 		if (!asyncpg) {
302 			(void) nfs4_waitfor_purge_complete(vp);
303 			nfs4_flush_pages(vp, cr);
304 		} else {
305 			pgflush_t *args;
306 
307 			/*
308 			 * We don't hold r_statelock while creating the
309 			 * thread, in case the call blocks.  So we use a
310 			 * flag to indicate that a page flush thread is
311 			 * active.
312 			 */
313 			mutex_enter(&rp->r_statelock);
314 			if (rp->r_flags & R4PGFLUSH) {
315 				mutex_exit(&rp->r_statelock);
316 			} else {
317 				rp->r_flags |= R4PGFLUSH;
318 				mutex_exit(&rp->r_statelock);
319 
320 				args = kmem_alloc(sizeof (pgflush_t),
321 				    KM_SLEEP);
322 				args->vp = vp;
323 				VN_HOLD(args->vp);
324 				args->cr = cr;
325 				crhold(args->cr);
326 				(void) zthread_create(NULL, 0,
327 				    nfs4_pgflush_thread, args, 0,
328 				    minclsyspri);
329 			}
330 		}
331 	}
332 
333 	/*
334 	 * Flush the readdir response cache.
335 	 */
336 	nfs4_purge_rddir_cache(vp);
337 }
338 
339 /*
340  * Invalidate all pages for the given file, after writing back the dirty
341  * ones.
342  */
343 
344 void
nfs4_flush_pages(vnode_t * vp,cred_t * cr)345 nfs4_flush_pages(vnode_t *vp, cred_t *cr)
346 {
347 	int error;
348 	rnode4_t *rp = VTOR4(vp);
349 
350 	error = VOP_PUTPAGE(vp, (u_offset_t)0, 0, B_INVAL, cr, NULL);
351 	if (error == ENOSPC || error == EDQUOT) {
352 		mutex_enter(&rp->r_statelock);
353 		if (!rp->r_error)
354 			rp->r_error = error;
355 		mutex_exit(&rp->r_statelock);
356 	}
357 }
358 
359 /*
360  * Page flush thread.
361  */
362 
363 static void
nfs4_pgflush_thread(pgflush_t * args)364 nfs4_pgflush_thread(pgflush_t *args)
365 {
366 	rnode4_t *rp = VTOR4(args->vp);
367 
368 	/* remember which thread we are, so we don't deadlock ourselves */
369 	mutex_enter(&rp->r_statelock);
370 	ASSERT(rp->r_pgflush == NULL);
371 	rp->r_pgflush = curthread;
372 	mutex_exit(&rp->r_statelock);
373 
374 	nfs4_flush_pages(args->vp, args->cr);
375 
376 	mutex_enter(&rp->r_statelock);
377 	rp->r_pgflush = NULL;
378 	rp->r_flags &= ~R4PGFLUSH;
379 	cv_broadcast(&rp->r_cv);
380 	mutex_exit(&rp->r_statelock);
381 
382 	VN_RELE(args->vp);
383 	crfree(args->cr);
384 	kmem_free(args, sizeof (pgflush_t));
385 	zthread_exit();
386 }
387 
388 /*
389  * Purge the readdir cache of all entries which are not currently
390  * being filled.
391  */
392 void
nfs4_purge_rddir_cache(vnode_t * vp)393 nfs4_purge_rddir_cache(vnode_t *vp)
394 {
395 	rnode4_t *rp;
396 
397 	rp = VTOR4(vp);
398 
399 	mutex_enter(&rp->r_statelock);
400 	rp->r_direof = NULL;
401 	rp->r_flags &= ~R4LOOKUP;
402 	rp->r_flags |= R4READDIRWATTR;
403 	rddir4_cache_purge(rp);
404 	mutex_exit(&rp->r_statelock);
405 }
406 
407 /*
408  * Set attributes cache for given vnode using virtual attributes.  There is
409  * no cache validation, but if the attributes are deemed to be stale, they
410  * are ignored.  This corresponds to nfs3_attrcache().
411  *
412  * Set the timeout value on the attribute cache and fill it
413  * with the passed in attributes.
414  */
415 void
nfs4_attrcache_noinval(vnode_t * vp,nfs4_ga_res_t * garp,hrtime_t t)416 nfs4_attrcache_noinval(vnode_t *vp, nfs4_ga_res_t *garp, hrtime_t t)
417 {
418 	rnode4_t *rp = VTOR4(vp);
419 
420 	mutex_enter(&rp->r_statelock);
421 	if (rp->r_time_attr_saved <= t)
422 		nfs4_attrcache_va(vp, garp, FALSE);
423 	mutex_exit(&rp->r_statelock);
424 }
425 
426 /*
427  * Use the passed in virtual attributes to check to see whether the
428  * data and metadata caches are valid, cache the new attributes, and
429  * then do the cache invalidation if required.
430  *
431  * The cache validation and caching of the new attributes is done
432  * atomically via the use of the mutex, r_statelock.  If required,
433  * the cache invalidation is done atomically w.r.t. the cache
434  * validation and caching of the attributes via the pseudo lock,
435  * r_serial.
436  *
437  * This routine is used to do cache validation and attributes caching
438  * for operations with a single set of post operation attributes.
439  */
440 
441 void
nfs4_attr_cache(vnode_t * vp,nfs4_ga_res_t * garp,hrtime_t t,cred_t * cr,int async,change_info4 * cinfo)442 nfs4_attr_cache(vnode_t *vp, nfs4_ga_res_t *garp,
443     hrtime_t t, cred_t *cr, int async,
444     change_info4 *cinfo)
445 {
446 	rnode4_t *rp;
447 	int mtime_changed = 0;
448 	int ctime_changed = 0;
449 	vsecattr_t *vsp;
450 	int was_serial, set_time_cache_inval, recov;
451 	vattr_t *vap = &garp->n4g_va;
452 	mntinfo4_t *mi = VTOMI4(vp);
453 	len_t preattr_rsize;
454 	boolean_t writemodify_set = B_FALSE;
455 	boolean_t cachepurge_set = B_FALSE;
456 
457 	ASSERT(mi->mi_vfsp->vfs_dev == garp->n4g_va.va_fsid);
458 
459 	/* Is curthread the recovery thread? */
460 	mutex_enter(&mi->mi_lock);
461 	recov = (VTOMI4(vp)->mi_recovthread == curthread);
462 	mutex_exit(&mi->mi_lock);
463 
464 	rp = VTOR4(vp);
465 	mutex_enter(&rp->r_statelock);
466 	was_serial = (rp->r_serial == curthread);
467 	if (rp->r_serial != NULL && !was_serial) {
468 		/*
469 		 * Purge current attrs and bail out to avoid potential deadlock
470 		 * between another thread caching attrs (r_serial thread), this
471 		 * thread, and a thread trying to read or write pages.
472 		 */
473 		PURGE_ATTRCACHE4_LOCKED(rp);
474 		mutex_exit(&rp->r_statelock);
475 		return;
476 	}
477 
478 	/*
479 	 * If there is a page flush thread, the current thread needs to
480 	 * bail out, to prevent a possible deadlock between the current
481 	 * thread (which might be in a start_op/end_op region), the
482 	 * recovery thread, and the page flush thread.  Expire the
483 	 * attribute cache, so that any attributes the current thread was
484 	 * going to set are not lost.
485 	 */
486 	if ((rp->r_flags & R4PGFLUSH) && rp->r_pgflush != curthread) {
487 		PURGE_ATTRCACHE4_LOCKED(rp);
488 		mutex_exit(&rp->r_statelock);
489 		return;
490 	}
491 
492 	if (rp->r_time_attr_saved > t) {
493 		/*
494 		 * Attributes have been cached since these attributes were
495 		 * probably made. If there is an inconsistency in what is
496 		 * cached, mark them invalid. If not, don't act on them.
497 		 */
498 		if (!CACHE4_VALID(rp, vap->va_mtime, vap->va_size))
499 			PURGE_ATTRCACHE4_LOCKED(rp);
500 		mutex_exit(&rp->r_statelock);
501 		return;
502 	}
503 	set_time_cache_inval = 0;
504 	if (cinfo) {
505 		/*
506 		 * Only directory modifying callers pass non-NULL cinfo.
507 		 */
508 		ASSERT(vp->v_type == VDIR);
509 		/*
510 		 * If the cache timeout either doesn't exist or hasn't expired,
511 		 * and dir didn't changed on server before dirmod op
512 		 * and dir didn't change after dirmod op but before getattr
513 		 * then there's a chance that the client's cached data for
514 		 * this object is current (not stale).  No immediate cache
515 		 * flush is required.
516 		 *
517 		 */
518 		if ((! rp->r_time_cache_inval || t < rp->r_time_cache_inval) &&
519 		    cinfo->before == rp->r_change &&
520 		    (garp->n4g_change_valid &&
521 		    cinfo->after == garp->n4g_change)) {
522 
523 			/*
524 			 * If atomic isn't set, then the before/after info
525 			 * cannot be blindly trusted.  For this case, we tell
526 			 * nfs4_attrcache_va to cache the attrs but also
527 			 * establish an absolute maximum cache timeout.  When
528 			 * the timeout is reached, caches will be flushed.
529 			 */
530 			if (! cinfo->atomic)
531 				set_time_cache_inval = 1;
532 		} else {
533 
534 			/*
535 			 * We're not sure exactly what changed, but we know
536 			 * what to do.  flush all caches for dir.  remove the
537 			 * attr timeout.
538 			 *
539 			 * a) timeout expired.  flush all caches.
540 			 * b) r_change != cinfo.before.  flush all caches.
541 			 * c) r_change == cinfo.before, but cinfo.after !=
542 			 *    post-op getattr(change).  flush all caches.
543 			 * d) post-op getattr(change) not provided by server.
544 			 *    flush all caches.
545 			 */
546 			mtime_changed = 1;
547 			ctime_changed = 1;
548 			rp->r_time_cache_inval = 0;
549 		}
550 	} else {
551 		/*
552 		 * Write thread after writing data to file on remote server,
553 		 * will always set R4WRITEMODIFIED to indicate that file on
554 		 * remote server was modified with a WRITE operation and would
555 		 * have marked attribute cache as timed out. If R4WRITEMODIFIED
556 		 * is set, then do not check for mtime and ctime change.
557 		 */
558 		if (!(rp->r_flags & R4WRITEMODIFIED)) {
559 			if (!CACHE4_VALID(rp, vap->va_mtime, vap->va_size))
560 				mtime_changed = 1;
561 
562 			if (rp->r_attr.va_ctime.tv_sec !=
563 			    vap->va_ctime.tv_sec ||
564 			    rp->r_attr.va_ctime.tv_nsec !=
565 			    vap->va_ctime.tv_nsec)
566 				ctime_changed = 1;
567 
568 			/*
569 			 * If the change attribute was not provided by server
570 			 * or it differs, then flush all caches.
571 			 */
572 			if (!garp->n4g_change_valid ||
573 			    rp->r_change != garp->n4g_change) {
574 				mtime_changed = 1;
575 				ctime_changed = 1;
576 			}
577 		} else {
578 			writemodify_set = B_TRUE;
579 		}
580 	}
581 
582 	preattr_rsize = rp->r_size;
583 
584 	nfs4_attrcache_va(vp, garp, set_time_cache_inval);
585 
586 	/*
587 	 * If we have updated filesize in nfs4_attrcache_va, as soon as we
588 	 * drop statelock we will be in transition of purging all
589 	 * our caches and updating them. It is possible for another
590 	 * thread to pick this new file size and read in zeroed data.
591 	 * stall other threads till cache purge is complete.
592 	 */
593 	if ((!cinfo) && (rp->r_size != preattr_rsize)) {
594 		/*
595 		 * If R4WRITEMODIFIED was set and we have updated the file
596 		 * size, Server's returned file size need not necessarily
597 		 * be because of this Client's WRITE. We need to purge
598 		 * all caches.
599 		 */
600 		if (writemodify_set)
601 			mtime_changed = 1;
602 
603 		if (mtime_changed && !(rp->r_flags & R4INCACHEPURGE)) {
604 			rp->r_flags |= R4INCACHEPURGE;
605 			cachepurge_set = B_TRUE;
606 		}
607 	}
608 
609 	if (!mtime_changed && !ctime_changed) {
610 		mutex_exit(&rp->r_statelock);
611 		return;
612 	}
613 
614 	rp->r_serial = curthread;
615 
616 	mutex_exit(&rp->r_statelock);
617 
618 	/*
619 	 * If we're the recov thread, then force async nfs4_purge_caches
620 	 * to avoid potential deadlock.
621 	 */
622 	if (mtime_changed)
623 		nfs4_purge_caches(vp, NFS4_NOPURGE_DNLC, cr, recov ? 1 : async);
624 
625 	if ((rp->r_flags & R4INCACHEPURGE) && cachepurge_set) {
626 		mutex_enter(&rp->r_statelock);
627 		rp->r_flags &= ~R4INCACHEPURGE;
628 		cv_broadcast(&rp->r_cv);
629 		mutex_exit(&rp->r_statelock);
630 		cachepurge_set = B_FALSE;
631 	}
632 
633 	if (ctime_changed) {
634 		(void) nfs4_access_purge_rp(rp);
635 		if (rp->r_secattr != NULL) {
636 			mutex_enter(&rp->r_statelock);
637 			vsp = rp->r_secattr;
638 			rp->r_secattr = NULL;
639 			mutex_exit(&rp->r_statelock);
640 			if (vsp != NULL)
641 				nfs4_acl_free_cache(vsp);
642 		}
643 	}
644 
645 	if (!was_serial) {
646 		mutex_enter(&rp->r_statelock);
647 		rp->r_serial = NULL;
648 		cv_broadcast(&rp->r_cv);
649 		mutex_exit(&rp->r_statelock);
650 	}
651 }
652 
653 /*
654  * Set attributes cache for given vnode using virtual attributes.
655  *
656  * Set the timeout value on the attribute cache and fill it
657  * with the passed in attributes.
658  *
659  * The caller must be holding r_statelock.
660  */
661 static void
nfs4_attrcache_va(vnode_t * vp,nfs4_ga_res_t * garp,int set_cache_timeout)662 nfs4_attrcache_va(vnode_t *vp, nfs4_ga_res_t *garp, int set_cache_timeout)
663 {
664 	rnode4_t *rp;
665 	mntinfo4_t *mi;
666 	hrtime_t delta;
667 	hrtime_t now;
668 	vattr_t *vap = &garp->n4g_va;
669 
670 	rp = VTOR4(vp);
671 
672 	ASSERT(MUTEX_HELD(&rp->r_statelock));
673 	ASSERT(vap->va_mask == AT_ALL);
674 
675 	/* Switch to master before checking v_flag */
676 	if (IS_SHADOW(vp, rp))
677 		vp = RTOV4(rp);
678 
679 	now = gethrtime();
680 
681 	mi = VTOMI4(vp);
682 
683 	/*
684 	 * Only establish a new cache timeout (if requested).  Never
685 	 * extend a timeout.  Never clear a timeout.  Clearing a timeout
686 	 * is done by nfs4_update_dircaches (ancestor in our call chain)
687 	 */
688 	if (set_cache_timeout && ! rp->r_time_cache_inval)
689 		rp->r_time_cache_inval = now + mi->mi_acdirmax;
690 
691 	/*
692 	 * Delta is the number of nanoseconds that we will
693 	 * cache the attributes of the file.  It is based on
694 	 * the number of nanoseconds since the last time that
695 	 * we detected a change.  The assumption is that files
696 	 * that changed recently are likely to change again.
697 	 * There is a minimum and a maximum for regular files
698 	 * and for directories which is enforced though.
699 	 *
700 	 * Using the time since last change was detected
701 	 * eliminates direct comparison or calculation
702 	 * using mixed client and server times.  NFS does
703 	 * not make any assumptions regarding the client
704 	 * and server clocks being synchronized.
705 	 */
706 	if (vap->va_mtime.tv_sec != rp->r_attr.va_mtime.tv_sec ||
707 	    vap->va_mtime.tv_nsec != rp->r_attr.va_mtime.tv_nsec ||
708 	    vap->va_size != rp->r_attr.va_size) {
709 		rp->r_time_attr_saved = now;
710 	}
711 
712 	if ((mi->mi_flags & MI4_NOAC) || (vp->v_flag & VNOCACHE))
713 		delta = 0;
714 	else {
715 		delta = now - rp->r_time_attr_saved;
716 		if (vp->v_type == VDIR) {
717 			if (delta < mi->mi_acdirmin)
718 				delta = mi->mi_acdirmin;
719 			else if (delta > mi->mi_acdirmax)
720 				delta = mi->mi_acdirmax;
721 		} else {
722 			if (delta < mi->mi_acregmin)
723 				delta = mi->mi_acregmin;
724 			else if (delta > mi->mi_acregmax)
725 				delta = mi->mi_acregmax;
726 		}
727 	}
728 	rp->r_time_attr_inval = now + delta;
729 
730 	rp->r_attr = *vap;
731 	if (garp->n4g_change_valid)
732 		rp->r_change = garp->n4g_change;
733 
734 	/*
735 	 * The attributes that were returned may be valid and can
736 	 * be used, but they may not be allowed to be cached.
737 	 * Reset the timers to cause immediate invalidation and
738 	 * clear r_change so no VERIFY operations will suceed
739 	 */
740 	if (garp->n4g_attrwhy == NFS4_GETATTR_NOCACHE_OK) {
741 		rp->r_time_attr_inval = now;
742 		rp->r_time_attr_saved = now;
743 		rp->r_change = 0;
744 	}
745 
746 	/*
747 	 * If mounted_on_fileid returned AND the object is a stub,
748 	 * then set object's va_nodeid to the mounted over fid
749 	 * returned by server.
750 	 *
751 	 * If mounted_on_fileid not provided/supported, then
752 	 * just set it to 0 for now.  Eventually it would be
753 	 * better to set it to a hashed version of FH.  This
754 	 * would probably be good enough to provide a unique
755 	 * fid/d_ino within a dir.
756 	 *
757 	 * We don't need to carry mounted_on_fileid in the
758 	 * rnode as long as the client never requests fileid
759 	 * without also requesting mounted_on_fileid.  For
760 	 * now, it stays.
761 	 */
762 	if (garp->n4g_mon_fid_valid) {
763 		rp->r_mntd_fid = garp->n4g_mon_fid;
764 
765 		if (RP_ISSTUB(rp))
766 			rp->r_attr.va_nodeid = rp->r_mntd_fid;
767 	}
768 
769 	/*
770 	 * Check to see if there are valid pathconf bits to
771 	 * cache in the rnode.
772 	 */
773 	if (garp->n4g_ext_res) {
774 		if (garp->n4g_ext_res->n4g_pc4.pc4_cache_valid) {
775 			rp->r_pathconf = garp->n4g_ext_res->n4g_pc4;
776 		} else {
777 			if (garp->n4g_ext_res->n4g_pc4.pc4_xattr_valid) {
778 				rp->r_pathconf.pc4_xattr_valid = TRUE;
779 				rp->r_pathconf.pc4_xattr_exists =
780 				    garp->n4g_ext_res->n4g_pc4.pc4_xattr_exists;
781 			}
782 		}
783 	}
784 	/*
785 	 * Update the size of the file if there is no cached data or if
786 	 * the cached data is clean and there is no data being written
787 	 * out.
788 	 */
789 	if (rp->r_size != vap->va_size &&
790 	    (!vn_has_cached_data(vp) ||
791 	    (!(rp->r_flags & R4DIRTY) && rp->r_count == 0))) {
792 		rp->r_size = vap->va_size;
793 	}
794 	nfs_setswaplike(vp, vap);
795 	rp->r_flags &= ~R4WRITEMODIFIED;
796 }
797 
798 /*
799  * Get attributes over-the-wire and update attributes cache
800  * if no error occurred in the over-the-wire operation.
801  * Return 0 if successful, otherwise error.
802  */
803 int
nfs4_getattr_otw(vnode_t * vp,nfs4_ga_res_t * garp,cred_t * cr,int get_acl)804 nfs4_getattr_otw(vnode_t *vp, nfs4_ga_res_t *garp, cred_t *cr, int get_acl)
805 {
806 	mntinfo4_t *mi = VTOMI4(vp);
807 	hrtime_t t;
808 	nfs4_recov_state_t recov_state;
809 	nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
810 
811 	recov_state.rs_flags = 0;
812 	recov_state.rs_num_retry_despite_err = 0;
813 
814 	/* Save the original mount point security flavor */
815 	(void) save_mnt_secinfo(mi->mi_curr_serv);
816 
817 recov_retry:
818 
819 	if ((e.error = nfs4_start_fop(mi, vp, NULL, OH_GETATTR,
820 	    &recov_state, NULL))) {
821 		(void) check_mnt_secinfo(mi->mi_curr_serv, vp);
822 		return (e.error);
823 	}
824 
825 	t = gethrtime();
826 
827 	nfs4_getattr_otw_norecovery(vp, garp, &e, cr, get_acl);
828 
829 	if (nfs4_needs_recovery(&e, FALSE, vp->v_vfsp)) {
830 		if (nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL,
831 		    NULL, OP_GETATTR, NULL, NULL, NULL) == FALSE)  {
832 			nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR,
833 			    &recov_state, 1);
834 			goto recov_retry;
835 		}
836 	}
837 
838 	nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, &recov_state, 0);
839 
840 	if (!e.error) {
841 		if (e.stat == NFS4_OK) {
842 			nfs4_attr_cache(vp, garp, t, cr, FALSE, NULL);
843 		} else {
844 			e.error = geterrno4(e.stat);
845 
846 			nfs4_purge_stale_fh(e.error, vp, cr);
847 		}
848 	}
849 
850 	/*
851 	 * If getattr a node that is a stub for a crossed
852 	 * mount point, keep the original secinfo flavor for
853 	 * the current file system, not the crossed one.
854 	 */
855 	(void) check_mnt_secinfo(mi->mi_curr_serv, vp);
856 
857 	return (e.error);
858 }
859 
860 /*
861  * Generate a compound to get attributes over-the-wire.
862  */
863 void
nfs4_getattr_otw_norecovery(vnode_t * vp,nfs4_ga_res_t * garp,nfs4_error_t * ep,cred_t * cr,int get_acl)864 nfs4_getattr_otw_norecovery(vnode_t *vp, nfs4_ga_res_t *garp,
865     nfs4_error_t *ep, cred_t *cr, int get_acl)
866 {
867 	COMPOUND4args_clnt args;
868 	COMPOUND4res_clnt res;
869 	int doqueue;
870 	rnode4_t *rp = VTOR4(vp);
871 	nfs_argop4 argop[2];
872 
873 	args.ctag = TAG_GETATTR;
874 
875 	args.array_len = 2;
876 	args.array = argop;
877 
878 	/* putfh */
879 	argop[0].argop = OP_CPUTFH;
880 	argop[0].nfs_argop4_u.opcputfh.sfh = rp->r_fh;
881 
882 	/* getattr */
883 	/*
884 	 * Unlike nfs version 2 and 3, where getattr returns all the
885 	 * attributes, nfs version 4 returns only the ones explicitly
886 	 * asked for. This creates problems, as some system functions
887 	 * (e.g. cache check) require certain attributes and if the
888 	 * cached node lacks some attributes such as uid/gid, it can
889 	 * affect system utilities (e.g. "ls") that rely on the information
890 	 * to be there. This can lead to anything from system crashes to
891 	 * corrupted information processed by user apps.
892 	 * So to ensure that all bases are covered, request at least
893 	 * the AT_ALL attribute mask.
894 	 */
895 	argop[1].argop = OP_GETATTR;
896 	argop[1].nfs_argop4_u.opgetattr.attr_request = NFS4_VATTR_MASK;
897 	if (get_acl)
898 		argop[1].nfs_argop4_u.opgetattr.attr_request |= FATTR4_ACL_MASK;
899 	argop[1].nfs_argop4_u.opgetattr.mi = VTOMI4(vp);
900 
901 	doqueue = 1;
902 
903 	rfs4call(VTOMI4(vp), &args, &res, cr, &doqueue, 0, ep);
904 
905 	if (ep->error)
906 		return;
907 
908 	if (res.status != NFS4_OK) {
909 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
910 		return;
911 	}
912 
913 	*garp = res.array[1].nfs_resop4_u.opgetattr.ga_res;
914 
915 	xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
916 }
917 
918 /*
919  * Return either cached or remote attributes. If get remote attr
920  * use them to check and invalidate caches, then cache the new attributes.
921  */
922 int
nfs4getattr(vnode_t * vp,vattr_t * vap,cred_t * cr)923 nfs4getattr(vnode_t *vp, vattr_t *vap, cred_t *cr)
924 {
925 	int error;
926 	rnode4_t *rp;
927 	nfs4_ga_res_t gar;
928 
929 	ASSERT(nfs4_consistent_type(vp));
930 
931 	/*
932 	 * If we've got cached attributes, we're done, otherwise go
933 	 * to the server to get attributes, which will update the cache
934 	 * in the process. Either way, use the cached attributes for
935 	 * the caller's vattr_t.
936 	 *
937 	 * Note that we ignore the gar set by the OTW call: the attr caching
938 	 * code may make adjustments when storing to the rnode, and we want
939 	 * to see those changes here.
940 	 */
941 	rp = VTOR4(vp);
942 	error = 0;
943 	mutex_enter(&rp->r_statelock);
944 	if (!ATTRCACHE4_VALID(vp)) {
945 		mutex_exit(&rp->r_statelock);
946 		error = nfs4_getattr_otw(vp, &gar, cr, 0);
947 		mutex_enter(&rp->r_statelock);
948 	}
949 
950 	if (!error)
951 		*vap = rp->r_attr;
952 
953 	/* Return the client's view of file size */
954 	vap->va_size = rp->r_size;
955 
956 	mutex_exit(&rp->r_statelock);
957 
958 	ASSERT(nfs4_consistent_type(vp));
959 
960 	return (error);
961 }
962 
963 int
nfs4_attr_otw(vnode_t * vp,nfs4_tag_type_t tag_type,nfs4_ga_res_t * garp,bitmap4 reqbitmap,cred_t * cr)964 nfs4_attr_otw(vnode_t *vp, nfs4_tag_type_t tag_type,
965     nfs4_ga_res_t *garp, bitmap4 reqbitmap, cred_t *cr)
966 {
967 	COMPOUND4args_clnt args;
968 	COMPOUND4res_clnt res;
969 	int doqueue;
970 	nfs_argop4 argop[2];
971 	mntinfo4_t *mi = VTOMI4(vp);
972 	bool_t needrecov = FALSE;
973 	nfs4_recov_state_t recov_state;
974 	nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
975 	nfs4_ga_ext_res_t *gerp;
976 
977 	recov_state.rs_flags = 0;
978 	recov_state.rs_num_retry_despite_err = 0;
979 
980 recov_retry:
981 	args.ctag = tag_type;
982 
983 	args.array_len = 2;
984 	args.array = argop;
985 
986 	e.error = nfs4_start_fop(mi, vp, NULL, OH_GETATTR, &recov_state, NULL);
987 	if (e.error)
988 		return (e.error);
989 
990 	/* putfh */
991 	argop[0].argop = OP_CPUTFH;
992 	argop[0].nfs_argop4_u.opcputfh.sfh = VTOR4(vp)->r_fh;
993 
994 	/* getattr */
995 	argop[1].argop = OP_GETATTR;
996 	argop[1].nfs_argop4_u.opgetattr.attr_request = reqbitmap;
997 	argop[1].nfs_argop4_u.opgetattr.mi = mi;
998 
999 	doqueue = 1;
1000 
1001 	NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
1002 	    "nfs4_attr_otw: %s call, rp %s", needrecov ? "recov" : "first",
1003 	    rnode4info(VTOR4(vp))));
1004 
1005 	rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
1006 
1007 	needrecov = nfs4_needs_recovery(&e, FALSE, vp->v_vfsp);
1008 	if (!needrecov && e.error) {
1009 		nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, &recov_state,
1010 		    needrecov);
1011 		return (e.error);
1012 	}
1013 
1014 	if (needrecov) {
1015 		bool_t abort;
1016 
1017 		NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
1018 		    "nfs4_attr_otw: initiating recovery\n"));
1019 
1020 		abort = nfs4_start_recovery(&e, VTOMI4(vp), vp, NULL, NULL,
1021 		    NULL, OP_GETATTR, NULL, NULL, NULL);
1022 		nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, &recov_state,
1023 		    needrecov);
1024 		if (!e.error) {
1025 			xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1026 			e.error = geterrno4(res.status);
1027 		}
1028 		if (abort == FALSE)
1029 			goto recov_retry;
1030 		return (e.error);
1031 	}
1032 
1033 	if (res.status) {
1034 		e.error = geterrno4(res.status);
1035 	} else {
1036 		gerp = garp->n4g_ext_res;
1037 		bcopy(&res.array[1].nfs_resop4_u.opgetattr.ga_res,
1038 		    garp, sizeof (nfs4_ga_res_t));
1039 		garp->n4g_ext_res = gerp;
1040 		if (garp->n4g_ext_res &&
1041 		    res.array[1].nfs_resop4_u.opgetattr.ga_res.n4g_ext_res)
1042 			bcopy(res.array[1].nfs_resop4_u.opgetattr.
1043 			    ga_res.n4g_ext_res,
1044 			    garp->n4g_ext_res, sizeof (nfs4_ga_ext_res_t));
1045 	}
1046 	xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
1047 	nfs4_end_fop(VTOMI4(vp), vp, NULL, OH_GETATTR, &recov_state,
1048 	    needrecov);
1049 	return (e.error);
1050 }
1051 
1052 /*
1053  * Asynchronous I/O parameters.  nfs_async_threads is the high-water mark
1054  * for the demand-based allocation of async threads per-mount.  The
1055  * nfs_async_timeout is the amount of time a thread will live after it
1056  * becomes idle, unless new I/O requests are received before the thread
1057  * dies.  See nfs4_async_putpage and nfs4_async_start.
1058  */
1059 
1060 static void	nfs4_async_start(struct vfs *);
1061 static void	nfs4_async_pgops_start(struct vfs *);
1062 static void	nfs4_async_common_start(struct vfs *, int);
1063 
1064 static void
free_async_args4(struct nfs4_async_reqs * args)1065 free_async_args4(struct nfs4_async_reqs *args)
1066 {
1067 	rnode4_t *rp;
1068 
1069 	if (args->a_io != NFS4_INACTIVE) {
1070 		rp = VTOR4(args->a_vp);
1071 		mutex_enter(&rp->r_statelock);
1072 		rp->r_count--;
1073 		if (args->a_io == NFS4_PUTAPAGE ||
1074 		    args->a_io == NFS4_PAGEIO)
1075 			rp->r_awcount--;
1076 		cv_broadcast(&rp->r_cv);
1077 		mutex_exit(&rp->r_statelock);
1078 		VN_RELE(args->a_vp);
1079 	}
1080 	crfree(args->a_cred);
1081 	kmem_free(args, sizeof (*args));
1082 }
1083 
1084 /*
1085  * Cross-zone thread creation and NFS access is disallowed, yet fsflush() and
1086  * pageout(), running in the global zone, have legitimate reasons to do
1087  * VOP_PUTPAGE(B_ASYNC) on other zones' NFS mounts.  We avoid the problem by
1088  * use of a a per-mount "asynchronous requests manager thread" which is
1089  * signaled by the various asynchronous work routines when there is
1090  * asynchronous work to be done.  It is responsible for creating new
1091  * worker threads if necessary, and notifying existing worker threads
1092  * that there is work to be done.
1093  *
1094  * In other words, it will "take the specifications from the customers and
1095  * give them to the engineers."
1096  *
1097  * Worker threads die off of their own accord if they are no longer
1098  * needed.
1099  *
1100  * This thread is killed when the zone is going away or the filesystem
1101  * is being unmounted.
1102  */
1103 void
nfs4_async_manager(vfs_t * vfsp)1104 nfs4_async_manager(vfs_t *vfsp)
1105 {
1106 	callb_cpr_t cprinfo;
1107 	mntinfo4_t *mi;
1108 	uint_t max_threads;
1109 
1110 	mi = VFTOMI4(vfsp);
1111 
1112 	CALLB_CPR_INIT(&cprinfo, &mi->mi_async_lock, callb_generic_cpr,
1113 	    "nfs4_async_manager");
1114 
1115 	mutex_enter(&mi->mi_async_lock);
1116 	/*
1117 	 * We want to stash the max number of threads that this mount was
1118 	 * allowed so we can use it later when the variable is set to zero as
1119 	 * part of the zone/mount going away.
1120 	 *
1121 	 * We want to be able to create at least one thread to handle
1122 	 * asynchronous inactive calls.
1123 	 */
1124 	max_threads = MAX(mi->mi_max_threads, 1);
1125 	/*
1126 	 * We don't want to wait for mi_max_threads to go to zero, since that
1127 	 * happens as part of a failed unmount, but this thread should only
1128 	 * exit when the mount is really going away.
1129 	 *
1130 	 * Once MI4_ASYNC_MGR_STOP is set, no more async operations will be
1131 	 * attempted: the various _async_*() functions know to do things
1132 	 * inline if mi_max_threads == 0.  Henceforth we just drain out the
1133 	 * outstanding requests.
1134 	 *
1135 	 * Note that we still create zthreads even if we notice the zone is
1136 	 * shutting down (MI4_ASYNC_MGR_STOP is set); this may cause the zone
1137 	 * shutdown sequence to take slightly longer in some cases, but
1138 	 * doesn't violate the protocol, as all threads will exit as soon as
1139 	 * they're done processing the remaining requests.
1140 	 */
1141 	for (;;) {
1142 		while (mi->mi_async_req_count > 0) {
1143 			/*
1144 			 * Paranoia: If the mount started out having
1145 			 * (mi->mi_max_threads == 0), and the value was
1146 			 * later changed (via a debugger or somesuch),
1147 			 * we could be confused since we will think we
1148 			 * can't create any threads, and the calling
1149 			 * code (which looks at the current value of
1150 			 * mi->mi_max_threads, now non-zero) thinks we
1151 			 * can.
1152 			 *
1153 			 * So, because we're paranoid, we create threads
1154 			 * up to the maximum of the original and the
1155 			 * current value. This means that future
1156 			 * (debugger-induced) alterations of
1157 			 * mi->mi_max_threads are ignored for our
1158 			 * purposes, but who told them they could change
1159 			 * random values on a live kernel anyhow?
1160 			 */
1161 			if (mi->mi_threads[NFS4_ASYNC_QUEUE] <
1162 			    MAX(mi->mi_max_threads, max_threads)) {
1163 				mi->mi_threads[NFS4_ASYNC_QUEUE]++;
1164 				mutex_exit(&mi->mi_async_lock);
1165 				MI4_HOLD(mi);
1166 				VFS_HOLD(vfsp);	/* hold for new thread */
1167 				(void) zthread_create(NULL, 0, nfs4_async_start,
1168 				    vfsp, 0, minclsyspri);
1169 				mutex_enter(&mi->mi_async_lock);
1170 			} else if (mi->mi_threads[NFS4_ASYNC_PGOPS_QUEUE] <
1171 			    NUM_ASYNC_PGOPS_THREADS) {
1172 				mi->mi_threads[NFS4_ASYNC_PGOPS_QUEUE]++;
1173 				mutex_exit(&mi->mi_async_lock);
1174 				MI4_HOLD(mi);
1175 				VFS_HOLD(vfsp); /* hold for new thread */
1176 				(void) zthread_create(NULL, 0,
1177 				    nfs4_async_pgops_start, vfsp, 0,
1178 				    minclsyspri);
1179 				mutex_enter(&mi->mi_async_lock);
1180 			}
1181 			NFS4_WAKE_ASYNC_WORKER(mi->mi_async_work_cv);
1182 			ASSERT(mi->mi_async_req_count != 0);
1183 			mi->mi_async_req_count--;
1184 		}
1185 
1186 		mutex_enter(&mi->mi_lock);
1187 		if (mi->mi_flags & MI4_ASYNC_MGR_STOP) {
1188 			mutex_exit(&mi->mi_lock);
1189 			break;
1190 		}
1191 		mutex_exit(&mi->mi_lock);
1192 
1193 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
1194 		cv_wait(&mi->mi_async_reqs_cv, &mi->mi_async_lock);
1195 		CALLB_CPR_SAFE_END(&cprinfo, &mi->mi_async_lock);
1196 	}
1197 
1198 	NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE,
1199 	    "nfs4_async_manager exiting for vfs %p\n", (void *)mi->mi_vfsp));
1200 	/*
1201 	 * Let everyone know we're done.
1202 	 */
1203 	mi->mi_manager_thread = NULL;
1204 	/*
1205 	 * Wake up the inactive thread.
1206 	 */
1207 	cv_broadcast(&mi->mi_inact_req_cv);
1208 	/*
1209 	 * Wake up anyone sitting in nfs4_async_manager_stop()
1210 	 */
1211 	cv_broadcast(&mi->mi_async_cv);
1212 	/*
1213 	 * There is no explicit call to mutex_exit(&mi->mi_async_lock)
1214 	 * since CALLB_CPR_EXIT is actually responsible for releasing
1215 	 * 'mi_async_lock'.
1216 	 */
1217 	CALLB_CPR_EXIT(&cprinfo);
1218 	VFS_RELE(vfsp);	/* release thread's hold */
1219 	MI4_RELE(mi);
1220 	zthread_exit();
1221 }
1222 
1223 /*
1224  * Signal (and wait for) the async manager thread to clean up and go away.
1225  */
1226 void
nfs4_async_manager_stop(vfs_t * vfsp)1227 nfs4_async_manager_stop(vfs_t *vfsp)
1228 {
1229 	mntinfo4_t *mi = VFTOMI4(vfsp);
1230 
1231 	mutex_enter(&mi->mi_async_lock);
1232 	mutex_enter(&mi->mi_lock);
1233 	mi->mi_flags |= MI4_ASYNC_MGR_STOP;
1234 	mutex_exit(&mi->mi_lock);
1235 	cv_broadcast(&mi->mi_async_reqs_cv);
1236 	/*
1237 	 * Wait for the async manager thread to die.
1238 	 */
1239 	while (mi->mi_manager_thread != NULL)
1240 		cv_wait(&mi->mi_async_cv, &mi->mi_async_lock);
1241 	mutex_exit(&mi->mi_async_lock);
1242 }
1243 
1244 int
nfs4_async_readahead(vnode_t * vp,u_offset_t blkoff,caddr_t addr,struct seg * seg,cred_t * cr,void (* readahead)(vnode_t *,u_offset_t,caddr_t,struct seg *,cred_t *))1245 nfs4_async_readahead(vnode_t *vp, u_offset_t blkoff, caddr_t addr,
1246     struct seg *seg, cred_t *cr, void (*readahead)(vnode_t *,
1247     u_offset_t, caddr_t, struct seg *, cred_t *))
1248 {
1249 	rnode4_t *rp;
1250 	mntinfo4_t *mi;
1251 	struct nfs4_async_reqs *args;
1252 
1253 	rp = VTOR4(vp);
1254 	ASSERT(rp->r_freef == NULL);
1255 
1256 	mi = VTOMI4(vp);
1257 
1258 	/*
1259 	 * If addr falls in a different segment, don't bother doing readahead.
1260 	 */
1261 	if (addr >= seg->s_base + seg->s_size)
1262 		return (-1);
1263 
1264 	/*
1265 	 * If we can't allocate a request structure, punt on the readahead.
1266 	 */
1267 	if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL)
1268 		return (-1);
1269 
1270 	/*
1271 	 * If a lock operation is pending, don't initiate any new
1272 	 * readaheads.  Otherwise, bump r_count to indicate the new
1273 	 * asynchronous I/O.
1274 	 */
1275 	if (!nfs_rw_tryenter(&rp->r_lkserlock, RW_READER)) {
1276 		kmem_free(args, sizeof (*args));
1277 		return (-1);
1278 	}
1279 	mutex_enter(&rp->r_statelock);
1280 	rp->r_count++;
1281 	mutex_exit(&rp->r_statelock);
1282 	nfs_rw_exit(&rp->r_lkserlock);
1283 
1284 	args->a_next = NULL;
1285 #ifdef DEBUG
1286 	args->a_queuer = curthread;
1287 #endif
1288 	VN_HOLD(vp);
1289 	args->a_vp = vp;
1290 	ASSERT(cr != NULL);
1291 	crhold(cr);
1292 	args->a_cred = cr;
1293 	args->a_io = NFS4_READ_AHEAD;
1294 	args->a_nfs4_readahead = readahead;
1295 	args->a_nfs4_blkoff = blkoff;
1296 	args->a_nfs4_seg = seg;
1297 	args->a_nfs4_addr = addr;
1298 
1299 	mutex_enter(&mi->mi_async_lock);
1300 
1301 	/*
1302 	 * If asyncio has been disabled, don't bother readahead.
1303 	 */
1304 	if (mi->mi_max_threads == 0) {
1305 		mutex_exit(&mi->mi_async_lock);
1306 		goto noasync;
1307 	}
1308 
1309 	/*
1310 	 * Link request structure into the async list and
1311 	 * wakeup async thread to do the i/o.
1312 	 */
1313 	if (mi->mi_async_reqs[NFS4_READ_AHEAD] == NULL) {
1314 		mi->mi_async_reqs[NFS4_READ_AHEAD] = args;
1315 		mi->mi_async_tail[NFS4_READ_AHEAD] = args;
1316 	} else {
1317 		mi->mi_async_tail[NFS4_READ_AHEAD]->a_next = args;
1318 		mi->mi_async_tail[NFS4_READ_AHEAD] = args;
1319 	}
1320 
1321 	if (mi->mi_io_kstats) {
1322 		mutex_enter(&mi->mi_lock);
1323 		kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
1324 		mutex_exit(&mi->mi_lock);
1325 	}
1326 
1327 	mi->mi_async_req_count++;
1328 	ASSERT(mi->mi_async_req_count != 0);
1329 	cv_signal(&mi->mi_async_reqs_cv);
1330 	mutex_exit(&mi->mi_async_lock);
1331 	return (0);
1332 
1333 noasync:
1334 	mutex_enter(&rp->r_statelock);
1335 	rp->r_count--;
1336 	cv_broadcast(&rp->r_cv);
1337 	mutex_exit(&rp->r_statelock);
1338 	VN_RELE(vp);
1339 	crfree(cr);
1340 	kmem_free(args, sizeof (*args));
1341 	return (-1);
1342 }
1343 
1344 static void
nfs4_async_start(struct vfs * vfsp)1345 nfs4_async_start(struct vfs *vfsp)
1346 {
1347 	nfs4_async_common_start(vfsp, NFS4_ASYNC_QUEUE);
1348 }
1349 
1350 static void
nfs4_async_pgops_start(struct vfs * vfsp)1351 nfs4_async_pgops_start(struct vfs *vfsp)
1352 {
1353 	nfs4_async_common_start(vfsp, NFS4_ASYNC_PGOPS_QUEUE);
1354 }
1355 
1356 /*
1357  * The async queues for each mounted file system are arranged as a
1358  * set of queues, one for each async i/o type.  Requests are taken
1359  * from the queues in a round-robin fashion.  A number of consecutive
1360  * requests are taken from each queue before moving on to the next
1361  * queue.  This functionality may allow the NFS Version 2 server to do
1362  * write clustering, even if the client is mixing writes and reads
1363  * because it will take multiple write requests from the queue
1364  * before processing any of the other async i/o types.
1365  *
1366  * XXX The nfs4_async_common_start thread is unsafe in the light of the present
1367  * model defined by cpr to suspend the system. Specifically over the
1368  * wire calls are cpr-unsafe. The thread should be reevaluated in
1369  * case of future updates to the cpr model.
1370  */
1371 static void
nfs4_async_common_start(struct vfs * vfsp,int async_queue)1372 nfs4_async_common_start(struct vfs *vfsp, int async_queue)
1373 {
1374 	struct nfs4_async_reqs *args;
1375 	mntinfo4_t *mi = VFTOMI4(vfsp);
1376 	clock_t time_left = 1;
1377 	callb_cpr_t cprinfo;
1378 	int i;
1379 	extern int nfs_async_timeout;
1380 	int async_types;
1381 	kcondvar_t *async_work_cv;
1382 
1383 	if (async_queue == NFS4_ASYNC_QUEUE) {
1384 		async_types = NFS4_ASYNC_TYPES;
1385 		async_work_cv = &mi->mi_async_work_cv[NFS4_ASYNC_QUEUE];
1386 	} else {
1387 		async_types = NFS4_ASYNC_PGOPS_TYPES;
1388 		async_work_cv = &mi->mi_async_work_cv[NFS4_ASYNC_PGOPS_QUEUE];
1389 	}
1390 
1391 	/*
1392 	 * Dynamic initialization of nfs_async_timeout to allow nfs to be
1393 	 * built in an implementation independent manner.
1394 	 */
1395 	if (nfs_async_timeout == -1)
1396 		nfs_async_timeout = NFS_ASYNC_TIMEOUT;
1397 
1398 	CALLB_CPR_INIT(&cprinfo, &mi->mi_async_lock, callb_generic_cpr, "nas");
1399 
1400 	mutex_enter(&mi->mi_async_lock);
1401 	for (;;) {
1402 		/*
1403 		 * Find the next queue containing an entry.  We start
1404 		 * at the current queue pointer and then round robin
1405 		 * through all of them until we either find a non-empty
1406 		 * queue or have looked through all of them.
1407 		 */
1408 		for (i = 0; i < async_types; i++) {
1409 			args = *mi->mi_async_curr[async_queue];
1410 			if (args != NULL)
1411 				break;
1412 			mi->mi_async_curr[async_queue]++;
1413 			if (mi->mi_async_curr[async_queue] ==
1414 			    &mi->mi_async_reqs[async_types]) {
1415 				mi->mi_async_curr[async_queue] =
1416 				    &mi->mi_async_reqs[0];
1417 			}
1418 		}
1419 		/*
1420 		 * If we didn't find a entry, then block until woken up
1421 		 * again and then look through the queues again.
1422 		 */
1423 		if (args == NULL) {
1424 			/*
1425 			 * Exiting is considered to be safe for CPR as well
1426 			 */
1427 			CALLB_CPR_SAFE_BEGIN(&cprinfo);
1428 
1429 			/*
1430 			 * Wakeup thread waiting to unmount the file
1431 			 * system only if all async threads are inactive.
1432 			 *
1433 			 * If we've timed-out and there's nothing to do,
1434 			 * then get rid of this thread.
1435 			 */
1436 			if (mi->mi_max_threads == 0 || time_left <= 0) {
1437 				--mi->mi_threads[async_queue];
1438 
1439 				if (mi->mi_threads[NFS4_ASYNC_QUEUE] == 0 &&
1440 				    mi->mi_threads[NFS4_ASYNC_PGOPS_QUEUE] == 0)
1441 					cv_signal(&mi->mi_async_cv);
1442 				CALLB_CPR_EXIT(&cprinfo);
1443 				VFS_RELE(vfsp);	/* release thread's hold */
1444 				MI4_RELE(mi);
1445 				zthread_exit();
1446 				/* NOTREACHED */
1447 			}
1448 			time_left = cv_reltimedwait(async_work_cv,
1449 			    &mi->mi_async_lock, nfs_async_timeout,
1450 			    TR_CLOCK_TICK);
1451 
1452 			CALLB_CPR_SAFE_END(&cprinfo, &mi->mi_async_lock);
1453 
1454 			continue;
1455 		} else {
1456 			time_left = 1;
1457 		}
1458 
1459 		/*
1460 		 * Remove the request from the async queue and then
1461 		 * update the current async request queue pointer.  If
1462 		 * the current queue is empty or we have removed enough
1463 		 * consecutive entries from it, then reset the counter
1464 		 * for this queue and then move the current pointer to
1465 		 * the next queue.
1466 		 */
1467 		*mi->mi_async_curr[async_queue] = args->a_next;
1468 		if (*mi->mi_async_curr[async_queue] == NULL ||
1469 		    --mi->mi_async_clusters[args->a_io] == 0) {
1470 			mi->mi_async_clusters[args->a_io] =
1471 			    mi->mi_async_init_clusters;
1472 			mi->mi_async_curr[async_queue]++;
1473 			if (mi->mi_async_curr[async_queue] ==
1474 			    &mi->mi_async_reqs[async_types]) {
1475 				mi->mi_async_curr[async_queue] =
1476 				    &mi->mi_async_reqs[0];
1477 			}
1478 		}
1479 
1480 		if (args->a_io != NFS4_INACTIVE && mi->mi_io_kstats) {
1481 			mutex_enter(&mi->mi_lock);
1482 			kstat_waitq_exit(KSTAT_IO_PTR(mi->mi_io_kstats));
1483 			mutex_exit(&mi->mi_lock);
1484 		}
1485 
1486 		mutex_exit(&mi->mi_async_lock);
1487 
1488 		/*
1489 		 * Obtain arguments from the async request structure.
1490 		 */
1491 		if (args->a_io == NFS4_READ_AHEAD && mi->mi_max_threads > 0) {
1492 			(*args->a_nfs4_readahead)(args->a_vp,
1493 			    args->a_nfs4_blkoff, args->a_nfs4_addr,
1494 			    args->a_nfs4_seg, args->a_cred);
1495 		} else if (args->a_io == NFS4_PUTAPAGE) {
1496 			(void) (*args->a_nfs4_putapage)(args->a_vp,
1497 			    args->a_nfs4_pp, args->a_nfs4_off,
1498 			    args->a_nfs4_len, args->a_nfs4_flags,
1499 			    args->a_cred);
1500 		} else if (args->a_io == NFS4_PAGEIO) {
1501 			(void) (*args->a_nfs4_pageio)(args->a_vp,
1502 			    args->a_nfs4_pp, args->a_nfs4_off,
1503 			    args->a_nfs4_len, args->a_nfs4_flags,
1504 			    args->a_cred);
1505 		} else if (args->a_io == NFS4_READDIR) {
1506 			(void) ((*args->a_nfs4_readdir)(args->a_vp,
1507 			    args->a_nfs4_rdc, args->a_cred));
1508 		} else if (args->a_io == NFS4_COMMIT) {
1509 			(*args->a_nfs4_commit)(args->a_vp, args->a_nfs4_plist,
1510 			    args->a_nfs4_offset, args->a_nfs4_count,
1511 			    args->a_cred);
1512 		} else if (args->a_io == NFS4_INACTIVE) {
1513 			nfs4_inactive_otw(args->a_vp, args->a_cred);
1514 		}
1515 
1516 		/*
1517 		 * Now, release the vnode and free the credentials
1518 		 * structure.
1519 		 */
1520 		free_async_args4(args);
1521 		/*
1522 		 * Reacquire the mutex because it will be needed above.
1523 		 */
1524 		mutex_enter(&mi->mi_async_lock);
1525 	}
1526 }
1527 
1528 /*
1529  * nfs4_inactive_thread - look for vnodes that need over-the-wire calls as
1530  * part of VOP_INACTIVE.
1531  */
1532 
1533 void
nfs4_inactive_thread(mntinfo4_t * mi)1534 nfs4_inactive_thread(mntinfo4_t *mi)
1535 {
1536 	struct nfs4_async_reqs *args;
1537 	callb_cpr_t cprinfo;
1538 	vfs_t *vfsp = mi->mi_vfsp;
1539 
1540 	CALLB_CPR_INIT(&cprinfo, &mi->mi_async_lock, callb_generic_cpr,
1541 	    "nfs4_inactive_thread");
1542 
1543 	for (;;) {
1544 		mutex_enter(&mi->mi_async_lock);
1545 		args = mi->mi_async_reqs[NFS4_INACTIVE];
1546 		if (args == NULL) {
1547 			mutex_enter(&mi->mi_lock);
1548 			/*
1549 			 * We don't want to exit until the async manager is done
1550 			 * with its work; hence the check for mi_manager_thread
1551 			 * being NULL.
1552 			 *
1553 			 * The async manager thread will cv_broadcast() on
1554 			 * mi_inact_req_cv when it's done, at which point we'll
1555 			 * wake up and exit.
1556 			 */
1557 			if (mi->mi_manager_thread == NULL)
1558 				goto die;
1559 			mi->mi_flags |= MI4_INACTIVE_IDLE;
1560 			mutex_exit(&mi->mi_lock);
1561 			cv_signal(&mi->mi_async_cv);
1562 			CALLB_CPR_SAFE_BEGIN(&cprinfo);
1563 			cv_wait(&mi->mi_inact_req_cv, &mi->mi_async_lock);
1564 			CALLB_CPR_SAFE_END(&cprinfo, &mi->mi_async_lock);
1565 			mutex_exit(&mi->mi_async_lock);
1566 		} else {
1567 			mutex_enter(&mi->mi_lock);
1568 			mi->mi_flags &= ~MI4_INACTIVE_IDLE;
1569 			mutex_exit(&mi->mi_lock);
1570 			mi->mi_async_reqs[NFS4_INACTIVE] = args->a_next;
1571 			mutex_exit(&mi->mi_async_lock);
1572 			nfs4_inactive_otw(args->a_vp, args->a_cred);
1573 			crfree(args->a_cred);
1574 			kmem_free(args, sizeof (*args));
1575 		}
1576 	}
1577 die:
1578 	mutex_exit(&mi->mi_lock);
1579 	mi->mi_inactive_thread = NULL;
1580 	cv_signal(&mi->mi_async_cv);
1581 
1582 	/*
1583 	 * There is no explicit call to mutex_exit(&mi->mi_async_lock) since
1584 	 * CALLB_CPR_EXIT is actually responsible for releasing 'mi_async_lock'.
1585 	 */
1586 	CALLB_CPR_EXIT(&cprinfo);
1587 
1588 	NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE,
1589 	    "nfs4_inactive_thread exiting for vfs %p\n", (void *)vfsp));
1590 
1591 	MI4_RELE(mi);
1592 	zthread_exit();
1593 	/* NOTREACHED */
1594 }
1595 
1596 /*
1597  * nfs_async_stop:
1598  * Wait for all outstanding putpage operations and the inactive thread to
1599  * complete; nfs4_async_stop_sig() without interruptibility.
1600  */
1601 void
nfs4_async_stop(struct vfs * vfsp)1602 nfs4_async_stop(struct vfs *vfsp)
1603 {
1604 	mntinfo4_t *mi = VFTOMI4(vfsp);
1605 
1606 	/*
1607 	 * Wait for all outstanding async operations to complete and for
1608 	 * worker threads to exit.
1609 	 */
1610 	mutex_enter(&mi->mi_async_lock);
1611 	mi->mi_max_threads = 0;
1612 	NFS4_WAKEALL_ASYNC_WORKERS(mi->mi_async_work_cv);
1613 	while (mi->mi_threads[NFS4_ASYNC_QUEUE] != 0 ||
1614 	    mi->mi_threads[NFS4_ASYNC_PGOPS_QUEUE] != 0)
1615 		cv_wait(&mi->mi_async_cv, &mi->mi_async_lock);
1616 
1617 	/*
1618 	 * Wait for the inactive thread to finish doing what it's doing.  It
1619 	 * won't exit until the last reference to the vfs_t goes away.
1620 	 */
1621 	if (mi->mi_inactive_thread != NULL) {
1622 		mutex_enter(&mi->mi_lock);
1623 		while (!(mi->mi_flags & MI4_INACTIVE_IDLE) ||
1624 		    (mi->mi_async_reqs[NFS4_INACTIVE] != NULL)) {
1625 			mutex_exit(&mi->mi_lock);
1626 			cv_wait(&mi->mi_async_cv, &mi->mi_async_lock);
1627 			mutex_enter(&mi->mi_lock);
1628 		}
1629 		mutex_exit(&mi->mi_lock);
1630 	}
1631 	mutex_exit(&mi->mi_async_lock);
1632 }
1633 
1634 /*
1635  * nfs_async_stop_sig:
1636  * Wait for all outstanding putpage operations and the inactive thread to
1637  * complete. If a signal is delivered we will abort and return non-zero;
1638  * otherwise return 0. Since this routine is called from nfs4_unmount, we
1639  * need to make it interruptible.
1640  */
1641 int
nfs4_async_stop_sig(struct vfs * vfsp)1642 nfs4_async_stop_sig(struct vfs *vfsp)
1643 {
1644 	mntinfo4_t *mi = VFTOMI4(vfsp);
1645 	ushort_t omax;
1646 	bool_t intr = FALSE;
1647 
1648 	/*
1649 	 * Wait for all outstanding putpage operations to complete and for
1650 	 * worker threads to exit.
1651 	 */
1652 	mutex_enter(&mi->mi_async_lock);
1653 	omax = mi->mi_max_threads;
1654 	mi->mi_max_threads = 0;
1655 	NFS4_WAKEALL_ASYNC_WORKERS(mi->mi_async_work_cv);
1656 	while (mi->mi_threads[NFS4_ASYNC_QUEUE] != 0 ||
1657 	    mi->mi_threads[NFS4_ASYNC_PGOPS_QUEUE] != 0) {
1658 		if (!cv_wait_sig(&mi->mi_async_cv, &mi->mi_async_lock)) {
1659 			intr = TRUE;
1660 			goto interrupted;
1661 		}
1662 	}
1663 
1664 	/*
1665 	 * Wait for the inactive thread to finish doing what it's doing.  It
1666 	 * won't exit until the a last reference to the vfs_t goes away.
1667 	 */
1668 	if (mi->mi_inactive_thread != NULL) {
1669 		mutex_enter(&mi->mi_lock);
1670 		while (!(mi->mi_flags & MI4_INACTIVE_IDLE) ||
1671 		    (mi->mi_async_reqs[NFS4_INACTIVE] != NULL)) {
1672 			mutex_exit(&mi->mi_lock);
1673 			if (!cv_wait_sig(&mi->mi_async_cv,
1674 			    &mi->mi_async_lock)) {
1675 				intr = TRUE;
1676 				goto interrupted;
1677 			}
1678 			mutex_enter(&mi->mi_lock);
1679 		}
1680 		mutex_exit(&mi->mi_lock);
1681 	}
1682 interrupted:
1683 	if (intr)
1684 		mi->mi_max_threads = omax;
1685 	mutex_exit(&mi->mi_async_lock);
1686 
1687 	return (intr);
1688 }
1689 
1690 int
nfs4_async_putapage(vnode_t * vp,page_t * pp,u_offset_t off,size_t len,int flags,cred_t * cr,int (* putapage)(vnode_t *,page_t *,u_offset_t,size_t,int,cred_t *))1691 nfs4_async_putapage(vnode_t *vp, page_t *pp, u_offset_t off, size_t len,
1692     int flags, cred_t *cr, int (*putapage)(vnode_t *, page_t *,
1693     u_offset_t, size_t, int, cred_t *))
1694 {
1695 	rnode4_t *rp;
1696 	mntinfo4_t *mi;
1697 	struct nfs4_async_reqs *args;
1698 
1699 	ASSERT(flags & B_ASYNC);
1700 	ASSERT(vp->v_vfsp != NULL);
1701 
1702 	rp = VTOR4(vp);
1703 	ASSERT(rp->r_count > 0);
1704 
1705 	mi = VTOMI4(vp);
1706 
1707 	/*
1708 	 * If we can't allocate a request structure, do the putpage
1709 	 * operation synchronously in this thread's context.
1710 	 */
1711 	if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL)
1712 		goto noasync;
1713 
1714 	args->a_next = NULL;
1715 #ifdef DEBUG
1716 	args->a_queuer = curthread;
1717 #endif
1718 	VN_HOLD(vp);
1719 	args->a_vp = vp;
1720 	ASSERT(cr != NULL);
1721 	crhold(cr);
1722 	args->a_cred = cr;
1723 	args->a_io = NFS4_PUTAPAGE;
1724 	args->a_nfs4_putapage = putapage;
1725 	args->a_nfs4_pp = pp;
1726 	args->a_nfs4_off = off;
1727 	args->a_nfs4_len = (uint_t)len;
1728 	args->a_nfs4_flags = flags;
1729 
1730 	mutex_enter(&mi->mi_async_lock);
1731 
1732 	/*
1733 	 * If asyncio has been disabled, then make a synchronous request.
1734 	 * This check is done a second time in case async io was diabled
1735 	 * while this thread was blocked waiting for memory pressure to
1736 	 * reduce or for the queue to drain.
1737 	 */
1738 	if (mi->mi_max_threads == 0) {
1739 		mutex_exit(&mi->mi_async_lock);
1740 
1741 		VN_RELE(vp);
1742 		crfree(cr);
1743 		kmem_free(args, sizeof (*args));
1744 		goto noasync;
1745 	}
1746 
1747 	/*
1748 	 * Link request structure into the async list and
1749 	 * wakeup async thread to do the i/o.
1750 	 */
1751 	if (mi->mi_async_reqs[NFS4_PUTAPAGE] == NULL) {
1752 		mi->mi_async_reqs[NFS4_PUTAPAGE] = args;
1753 		mi->mi_async_tail[NFS4_PUTAPAGE] = args;
1754 	} else {
1755 		mi->mi_async_tail[NFS4_PUTAPAGE]->a_next = args;
1756 		mi->mi_async_tail[NFS4_PUTAPAGE] = args;
1757 	}
1758 
1759 	mutex_enter(&rp->r_statelock);
1760 	rp->r_count++;
1761 	rp->r_awcount++;
1762 	mutex_exit(&rp->r_statelock);
1763 
1764 	if (mi->mi_io_kstats) {
1765 		mutex_enter(&mi->mi_lock);
1766 		kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
1767 		mutex_exit(&mi->mi_lock);
1768 	}
1769 
1770 	mi->mi_async_req_count++;
1771 	ASSERT(mi->mi_async_req_count != 0);
1772 	cv_signal(&mi->mi_async_reqs_cv);
1773 	mutex_exit(&mi->mi_async_lock);
1774 	return (0);
1775 
1776 noasync:
1777 
1778 	if (curproc == proc_pageout || curproc == proc_fsflush) {
1779 		/*
1780 		 * If we get here in the context of the pageout/fsflush,
1781 		 * or we have run out of memory or we're attempting to
1782 		 * unmount we refuse to do a sync write, because this may
1783 		 * hang pageout/fsflush and the machine. In this case,
1784 		 * we just re-mark the page as dirty and punt on the page.
1785 		 *
1786 		 * Make sure B_FORCE isn't set.  We can re-mark the
1787 		 * pages as dirty and unlock the pages in one swoop by
1788 		 * passing in B_ERROR to pvn_write_done().  However,
1789 		 * we should make sure B_FORCE isn't set - we don't
1790 		 * want the page tossed before it gets written out.
1791 		 */
1792 		if (flags & B_FORCE)
1793 			flags &= ~(B_INVAL | B_FORCE);
1794 		pvn_write_done(pp, flags | B_ERROR);
1795 		return (0);
1796 	}
1797 
1798 	if (nfs_zone() != mi->mi_zone) {
1799 		/*
1800 		 * So this was a cross-zone sync putpage.
1801 		 *
1802 		 * We pass in B_ERROR to pvn_write_done() to re-mark the pages
1803 		 * as dirty and unlock them.
1804 		 *
1805 		 * We don't want to clear B_FORCE here as the caller presumably
1806 		 * knows what they're doing if they set it.
1807 		 */
1808 		pvn_write_done(pp, flags | B_ERROR);
1809 		return (EPERM);
1810 	}
1811 	return ((*putapage)(vp, pp, off, len, flags, cr));
1812 }
1813 
1814 int
nfs4_async_pageio(vnode_t * vp,page_t * pp,u_offset_t io_off,size_t io_len,int flags,cred_t * cr,int (* pageio)(vnode_t *,page_t *,u_offset_t,size_t,int,cred_t *))1815 nfs4_async_pageio(vnode_t *vp, page_t *pp, u_offset_t io_off, size_t io_len,
1816     int flags, cred_t *cr, int (*pageio)(vnode_t *, page_t *, u_offset_t,
1817     size_t, int, cred_t *))
1818 {
1819 	rnode4_t *rp;
1820 	mntinfo4_t *mi;
1821 	struct nfs4_async_reqs *args;
1822 
1823 	ASSERT(flags & B_ASYNC);
1824 	ASSERT(vp->v_vfsp != NULL);
1825 
1826 	rp = VTOR4(vp);
1827 	ASSERT(rp->r_count > 0);
1828 
1829 	mi = VTOMI4(vp);
1830 
1831 	/*
1832 	 * If we can't allocate a request structure, do the pageio
1833 	 * request synchronously in this thread's context.
1834 	 */
1835 	if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL)
1836 		goto noasync;
1837 
1838 	args->a_next = NULL;
1839 #ifdef DEBUG
1840 	args->a_queuer = curthread;
1841 #endif
1842 	VN_HOLD(vp);
1843 	args->a_vp = vp;
1844 	ASSERT(cr != NULL);
1845 	crhold(cr);
1846 	args->a_cred = cr;
1847 	args->a_io = NFS4_PAGEIO;
1848 	args->a_nfs4_pageio = pageio;
1849 	args->a_nfs4_pp = pp;
1850 	args->a_nfs4_off = io_off;
1851 	args->a_nfs4_len = (uint_t)io_len;
1852 	args->a_nfs4_flags = flags;
1853 
1854 	mutex_enter(&mi->mi_async_lock);
1855 
1856 	/*
1857 	 * If asyncio has been disabled, then make a synchronous request.
1858 	 * This check is done a second time in case async io was diabled
1859 	 * while this thread was blocked waiting for memory pressure to
1860 	 * reduce or for the queue to drain.
1861 	 */
1862 	if (mi->mi_max_threads == 0) {
1863 		mutex_exit(&mi->mi_async_lock);
1864 
1865 		VN_RELE(vp);
1866 		crfree(cr);
1867 		kmem_free(args, sizeof (*args));
1868 		goto noasync;
1869 	}
1870 
1871 	/*
1872 	 * Link request structure into the async list and
1873 	 * wakeup async thread to do the i/o.
1874 	 */
1875 	if (mi->mi_async_reqs[NFS4_PAGEIO] == NULL) {
1876 		mi->mi_async_reqs[NFS4_PAGEIO] = args;
1877 		mi->mi_async_tail[NFS4_PAGEIO] = args;
1878 	} else {
1879 		mi->mi_async_tail[NFS4_PAGEIO]->a_next = args;
1880 		mi->mi_async_tail[NFS4_PAGEIO] = args;
1881 	}
1882 
1883 	mutex_enter(&rp->r_statelock);
1884 	rp->r_count++;
1885 	rp->r_awcount++;
1886 	mutex_exit(&rp->r_statelock);
1887 
1888 	if (mi->mi_io_kstats) {
1889 		mutex_enter(&mi->mi_lock);
1890 		kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
1891 		mutex_exit(&mi->mi_lock);
1892 	}
1893 
1894 	mi->mi_async_req_count++;
1895 	ASSERT(mi->mi_async_req_count != 0);
1896 	cv_signal(&mi->mi_async_reqs_cv);
1897 	mutex_exit(&mi->mi_async_lock);
1898 	return (0);
1899 
1900 noasync:
1901 	/*
1902 	 * If we can't do it ASYNC, for reads we do nothing (but cleanup
1903 	 * the page list), for writes we do it synchronously, except for
1904 	 * proc_pageout/proc_fsflush as described below.
1905 	 */
1906 	if (flags & B_READ) {
1907 		pvn_read_done(pp, flags | B_ERROR);
1908 		return (0);
1909 	}
1910 
1911 	if (curproc == proc_pageout || curproc == proc_fsflush) {
1912 		/*
1913 		 * If we get here in the context of the pageout/fsflush,
1914 		 * we refuse to do a sync write, because this may hang
1915 		 * pageout/fsflush (and the machine). In this case, we just
1916 		 * re-mark the page as dirty and punt on the page.
1917 		 *
1918 		 * Make sure B_FORCE isn't set.  We can re-mark the
1919 		 * pages as dirty and unlock the pages in one swoop by
1920 		 * passing in B_ERROR to pvn_write_done().  However,
1921 		 * we should make sure B_FORCE isn't set - we don't
1922 		 * want the page tossed before it gets written out.
1923 		 */
1924 		if (flags & B_FORCE)
1925 			flags &= ~(B_INVAL | B_FORCE);
1926 		pvn_write_done(pp, flags | B_ERROR);
1927 		return (0);
1928 	}
1929 
1930 	if (nfs_zone() != mi->mi_zone) {
1931 		/*
1932 		 * So this was a cross-zone sync pageio.  We pass in B_ERROR
1933 		 * to pvn_write_done() to re-mark the pages as dirty and unlock
1934 		 * them.
1935 		 *
1936 		 * We don't want to clear B_FORCE here as the caller presumably
1937 		 * knows what they're doing if they set it.
1938 		 */
1939 		pvn_write_done(pp, flags | B_ERROR);
1940 		return (EPERM);
1941 	}
1942 	return ((*pageio)(vp, pp, io_off, io_len, flags, cr));
1943 }
1944 
1945 void
nfs4_async_readdir(vnode_t * vp,rddir4_cache * rdc,cred_t * cr,int (* readdir)(vnode_t *,rddir4_cache *,cred_t *))1946 nfs4_async_readdir(vnode_t *vp, rddir4_cache *rdc, cred_t *cr,
1947     int (*readdir)(vnode_t *, rddir4_cache *, cred_t *))
1948 {
1949 	rnode4_t *rp;
1950 	mntinfo4_t *mi;
1951 	struct nfs4_async_reqs *args;
1952 
1953 	rp = VTOR4(vp);
1954 	ASSERT(rp->r_freef == NULL);
1955 
1956 	mi = VTOMI4(vp);
1957 
1958 	/*
1959 	 * If we can't allocate a request structure, skip the readdir.
1960 	 */
1961 	if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL)
1962 		goto noasync;
1963 
1964 	args->a_next = NULL;
1965 #ifdef DEBUG
1966 	args->a_queuer = curthread;
1967 #endif
1968 	VN_HOLD(vp);
1969 	args->a_vp = vp;
1970 	ASSERT(cr != NULL);
1971 	crhold(cr);
1972 	args->a_cred = cr;
1973 	args->a_io = NFS4_READDIR;
1974 	args->a_nfs4_readdir = readdir;
1975 	args->a_nfs4_rdc = rdc;
1976 
1977 	mutex_enter(&mi->mi_async_lock);
1978 
1979 	/*
1980 	 * If asyncio has been disabled, then skip this request
1981 	 */
1982 	if (mi->mi_max_threads == 0) {
1983 		mutex_exit(&mi->mi_async_lock);
1984 
1985 		VN_RELE(vp);
1986 		crfree(cr);
1987 		kmem_free(args, sizeof (*args));
1988 		goto noasync;
1989 	}
1990 
1991 	/*
1992 	 * Link request structure into the async list and
1993 	 * wakeup async thread to do the i/o.
1994 	 */
1995 	if (mi->mi_async_reqs[NFS4_READDIR] == NULL) {
1996 		mi->mi_async_reqs[NFS4_READDIR] = args;
1997 		mi->mi_async_tail[NFS4_READDIR] = args;
1998 	} else {
1999 		mi->mi_async_tail[NFS4_READDIR]->a_next = args;
2000 		mi->mi_async_tail[NFS4_READDIR] = args;
2001 	}
2002 
2003 	mutex_enter(&rp->r_statelock);
2004 	rp->r_count++;
2005 	mutex_exit(&rp->r_statelock);
2006 
2007 	if (mi->mi_io_kstats) {
2008 		mutex_enter(&mi->mi_lock);
2009 		kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
2010 		mutex_exit(&mi->mi_lock);
2011 	}
2012 
2013 	mi->mi_async_req_count++;
2014 	ASSERT(mi->mi_async_req_count != 0);
2015 	cv_signal(&mi->mi_async_reqs_cv);
2016 	mutex_exit(&mi->mi_async_lock);
2017 	return;
2018 
2019 noasync:
2020 	mutex_enter(&rp->r_statelock);
2021 	rdc->entries = NULL;
2022 	/*
2023 	 * Indicate that no one is trying to fill this entry and
2024 	 * it still needs to be filled.
2025 	 */
2026 	rdc->flags &= ~RDDIR;
2027 	rdc->flags |= RDDIRREQ;
2028 	rddir4_cache_rele(rp, rdc);
2029 	mutex_exit(&rp->r_statelock);
2030 }
2031 
2032 void
nfs4_async_commit(vnode_t * vp,page_t * plist,offset3 offset,count3 count,cred_t * cr,void (* commit)(vnode_t *,page_t *,offset3,count3,cred_t *))2033 nfs4_async_commit(vnode_t *vp, page_t *plist, offset3 offset, count3 count,
2034     cred_t *cr, void (*commit)(vnode_t *, page_t *, offset3, count3,
2035     cred_t *))
2036 {
2037 	rnode4_t *rp;
2038 	mntinfo4_t *mi;
2039 	struct nfs4_async_reqs *args;
2040 	page_t *pp;
2041 
2042 	rp = VTOR4(vp);
2043 	mi = VTOMI4(vp);
2044 
2045 	/*
2046 	 * If we can't allocate a request structure, do the commit
2047 	 * operation synchronously in this thread's context.
2048 	 */
2049 	if ((args = kmem_alloc(sizeof (*args), KM_NOSLEEP)) == NULL)
2050 		goto noasync;
2051 
2052 	args->a_next = NULL;
2053 #ifdef DEBUG
2054 	args->a_queuer = curthread;
2055 #endif
2056 	VN_HOLD(vp);
2057 	args->a_vp = vp;
2058 	ASSERT(cr != NULL);
2059 	crhold(cr);
2060 	args->a_cred = cr;
2061 	args->a_io = NFS4_COMMIT;
2062 	args->a_nfs4_commit = commit;
2063 	args->a_nfs4_plist = plist;
2064 	args->a_nfs4_offset = offset;
2065 	args->a_nfs4_count = count;
2066 
2067 	mutex_enter(&mi->mi_async_lock);
2068 
2069 	/*
2070 	 * If asyncio has been disabled, then make a synchronous request.
2071 	 * This check is done a second time in case async io was diabled
2072 	 * while this thread was blocked waiting for memory pressure to
2073 	 * reduce or for the queue to drain.
2074 	 */
2075 	if (mi->mi_max_threads == 0) {
2076 		mutex_exit(&mi->mi_async_lock);
2077 
2078 		VN_RELE(vp);
2079 		crfree(cr);
2080 		kmem_free(args, sizeof (*args));
2081 		goto noasync;
2082 	}
2083 
2084 	/*
2085 	 * Link request structure into the async list and
2086 	 * wakeup async thread to do the i/o.
2087 	 */
2088 	if (mi->mi_async_reqs[NFS4_COMMIT] == NULL) {
2089 		mi->mi_async_reqs[NFS4_COMMIT] = args;
2090 		mi->mi_async_tail[NFS4_COMMIT] = args;
2091 	} else {
2092 		mi->mi_async_tail[NFS4_COMMIT]->a_next = args;
2093 		mi->mi_async_tail[NFS4_COMMIT] = args;
2094 	}
2095 
2096 	mutex_enter(&rp->r_statelock);
2097 	rp->r_count++;
2098 	mutex_exit(&rp->r_statelock);
2099 
2100 	if (mi->mi_io_kstats) {
2101 		mutex_enter(&mi->mi_lock);
2102 		kstat_waitq_enter(KSTAT_IO_PTR(mi->mi_io_kstats));
2103 		mutex_exit(&mi->mi_lock);
2104 	}
2105 
2106 	mi->mi_async_req_count++;
2107 	ASSERT(mi->mi_async_req_count != 0);
2108 	cv_signal(&mi->mi_async_reqs_cv);
2109 	mutex_exit(&mi->mi_async_lock);
2110 	return;
2111 
2112 noasync:
2113 	if (curproc == proc_pageout || curproc == proc_fsflush ||
2114 	    nfs_zone() != mi->mi_zone) {
2115 		while (plist != NULL) {
2116 			pp = plist;
2117 			page_sub(&plist, pp);
2118 			pp->p_fsdata = C_COMMIT;
2119 			page_unlock(pp);
2120 		}
2121 		return;
2122 	}
2123 	(*commit)(vp, plist, offset, count, cr);
2124 }
2125 
2126 /*
2127  * nfs4_async_inactive - hand off a VOP_INACTIVE call to a thread.  The
2128  * reference to the vnode is handed over to the thread; the caller should
2129  * no longer refer to the vnode.
2130  *
2131  * Unlike most of the async routines, this handoff is needed for
2132  * correctness reasons, not just performance.  So doing operations in the
2133  * context of the current thread is not an option.
2134  */
2135 void
nfs4_async_inactive(vnode_t * vp,cred_t * cr)2136 nfs4_async_inactive(vnode_t *vp, cred_t *cr)
2137 {
2138 	mntinfo4_t *mi;
2139 	struct nfs4_async_reqs *args;
2140 	boolean_t signal_inactive_thread = B_FALSE;
2141 
2142 	mi = VTOMI4(vp);
2143 
2144 	args = kmem_alloc(sizeof (*args), KM_SLEEP);
2145 	args->a_next = NULL;
2146 #ifdef DEBUG
2147 	args->a_queuer = curthread;
2148 #endif
2149 	args->a_vp = vp;
2150 	ASSERT(cr != NULL);
2151 	crhold(cr);
2152 	args->a_cred = cr;
2153 	args->a_io = NFS4_INACTIVE;
2154 
2155 	/*
2156 	 * Note that we don't check mi->mi_max_threads here, since we
2157 	 * *need* to get rid of this vnode regardless of whether someone
2158 	 * set nfs4_max_threads to zero in /etc/system.
2159 	 *
2160 	 * The manager thread knows about this and is willing to create
2161 	 * at least one thread to accommodate us.
2162 	 */
2163 	mutex_enter(&mi->mi_async_lock);
2164 	if (mi->mi_inactive_thread == NULL) {
2165 		rnode4_t *rp;
2166 		vnode_t *unldvp = NULL;
2167 		char *unlname;
2168 		cred_t *unlcred;
2169 
2170 		mutex_exit(&mi->mi_async_lock);
2171 		/*
2172 		 * We just need to free up the memory associated with the
2173 		 * vnode, which can be safely done from within the current
2174 		 * context.
2175 		 */
2176 		crfree(cr);	/* drop our reference */
2177 		kmem_free(args, sizeof (*args));
2178 		rp = VTOR4(vp);
2179 		mutex_enter(&rp->r_statelock);
2180 		if (rp->r_unldvp != NULL) {
2181 			unldvp = rp->r_unldvp;
2182 			rp->r_unldvp = NULL;
2183 			unlname = rp->r_unlname;
2184 			rp->r_unlname = NULL;
2185 			unlcred = rp->r_unlcred;
2186 			rp->r_unlcred = NULL;
2187 		}
2188 		mutex_exit(&rp->r_statelock);
2189 		/*
2190 		 * No need to explicitly throw away any cached pages.  The
2191 		 * eventual r4inactive() will attempt a synchronous
2192 		 * VOP_PUTPAGE() which will immediately fail since the request
2193 		 * is coming from the wrong zone, and then will proceed to call
2194 		 * nfs4_invalidate_pages() which will clean things up for us.
2195 		 *
2196 		 * Throw away the delegation here so rp4_addfree()'s attempt to
2197 		 * return any existing delegations becomes a no-op.
2198 		 */
2199 		if (rp->r_deleg_type != OPEN_DELEGATE_NONE) {
2200 			(void) nfs_rw_enter_sig(&mi->mi_recovlock, RW_READER,
2201 			    FALSE);
2202 			(void) nfs4delegreturn(rp, NFS4_DR_DISCARD);
2203 			nfs_rw_exit(&mi->mi_recovlock);
2204 		}
2205 		nfs4_clear_open_streams(rp);
2206 
2207 		rp4_addfree(rp, cr);
2208 		if (unldvp != NULL) {
2209 			kmem_free(unlname, MAXNAMELEN);
2210 			VN_RELE(unldvp);
2211 			crfree(unlcred);
2212 		}
2213 		return;
2214 	}
2215 
2216 	if (mi->mi_manager_thread == NULL) {
2217 		/*
2218 		 * We want to talk to the inactive thread.
2219 		 */
2220 		signal_inactive_thread = B_TRUE;
2221 	}
2222 
2223 	/*
2224 	 * Enqueue the vnode and wake up either the special thread (empty
2225 	 * list) or an async thread.
2226 	 */
2227 	if (mi->mi_async_reqs[NFS4_INACTIVE] == NULL) {
2228 		mi->mi_async_reqs[NFS4_INACTIVE] = args;
2229 		mi->mi_async_tail[NFS4_INACTIVE] = args;
2230 		signal_inactive_thread = B_TRUE;
2231 	} else {
2232 		mi->mi_async_tail[NFS4_INACTIVE]->a_next = args;
2233 		mi->mi_async_tail[NFS4_INACTIVE] = args;
2234 	}
2235 	if (signal_inactive_thread) {
2236 		cv_signal(&mi->mi_inact_req_cv);
2237 	} else  {
2238 		mi->mi_async_req_count++;
2239 		ASSERT(mi->mi_async_req_count != 0);
2240 		cv_signal(&mi->mi_async_reqs_cv);
2241 	}
2242 
2243 	mutex_exit(&mi->mi_async_lock);
2244 }
2245 
2246 int
writerp4(rnode4_t * rp,caddr_t base,int tcount,struct uio * uio,int pgcreated)2247 writerp4(rnode4_t *rp, caddr_t base, int tcount, struct uio *uio, int pgcreated)
2248 {
2249 	int pagecreate;
2250 	int n;
2251 	int saved_n;
2252 	caddr_t saved_base;
2253 	u_offset_t offset;
2254 	int error;
2255 	int sm_error;
2256 	vnode_t *vp = RTOV(rp);
2257 
2258 	ASSERT(tcount <= MAXBSIZE && tcount <= uio->uio_resid);
2259 	ASSERT(nfs_rw_lock_held(&rp->r_rwlock, RW_WRITER));
2260 	if (!vpm_enable) {
2261 		ASSERT(((uintptr_t)base & MAXBOFFSET) + tcount <= MAXBSIZE);
2262 	}
2263 
2264 	/*
2265 	 * Move bytes in at most PAGESIZE chunks. We must avoid
2266 	 * spanning pages in uiomove() because page faults may cause
2267 	 * the cache to be invalidated out from under us. The r_size is not
2268 	 * updated until after the uiomove. If we push the last page of a
2269 	 * file before r_size is correct, we will lose the data written past
2270 	 * the current (and invalid) r_size.
2271 	 */
2272 	do {
2273 		offset = uio->uio_loffset;
2274 		pagecreate = 0;
2275 
2276 		/*
2277 		 * n is the number of bytes required to satisfy the request
2278 		 *   or the number of bytes to fill out the page.
2279 		 */
2280 		n = (int)MIN((PAGESIZE - (offset & PAGEOFFSET)), tcount);
2281 
2282 		/*
2283 		 * Check to see if we can skip reading in the page
2284 		 * and just allocate the memory.  We can do this
2285 		 * if we are going to rewrite the entire mapping
2286 		 * or if we are going to write to or beyond the current
2287 		 * end of file from the beginning of the mapping.
2288 		 *
2289 		 * The read of r_size is now protected by r_statelock.
2290 		 */
2291 		mutex_enter(&rp->r_statelock);
2292 		/*
2293 		 * When pgcreated is nonzero the caller has already done
2294 		 * a segmap_getmapflt with forcefault 0 and S_WRITE. With
2295 		 * segkpm this means we already have at least one page
2296 		 * created and mapped at base.
2297 		 */
2298 		pagecreate = pgcreated ||
2299 		    ((offset & PAGEOFFSET) == 0 &&
2300 		    (n == PAGESIZE || ((offset + n) >= rp->r_size)));
2301 
2302 		mutex_exit(&rp->r_statelock);
2303 
2304 		if (!vpm_enable && pagecreate) {
2305 			/*
2306 			 * The last argument tells segmap_pagecreate() to
2307 			 * always lock the page, as opposed to sometimes
2308 			 * returning with the page locked. This way we avoid a
2309 			 * fault on the ensuing uiomove(), but also
2310 			 * more importantly (to fix bug 1094402) we can
2311 			 * call segmap_fault() to unlock the page in all
2312 			 * cases. An alternative would be to modify
2313 			 * segmap_pagecreate() to tell us when it is
2314 			 * locking a page, but that's a fairly major
2315 			 * interface change.
2316 			 */
2317 			if (pgcreated == 0)
2318 				(void) segmap_pagecreate(segkmap, base,
2319 				    (uint_t)n, 1);
2320 			saved_base = base;
2321 			saved_n = n;
2322 		}
2323 
2324 		/*
2325 		 * The number of bytes of data in the last page can not
2326 		 * be accurately be determined while page is being
2327 		 * uiomove'd to and the size of the file being updated.
2328 		 * Thus, inform threads which need to know accurately
2329 		 * how much data is in the last page of the file.  They
2330 		 * will not do the i/o immediately, but will arrange for
2331 		 * the i/o to happen later when this modify operation
2332 		 * will have finished.
2333 		 */
2334 		ASSERT(!(rp->r_flags & R4MODINPROGRESS));
2335 		mutex_enter(&rp->r_statelock);
2336 		rp->r_flags |= R4MODINPROGRESS;
2337 		rp->r_modaddr = (offset & MAXBMASK);
2338 		mutex_exit(&rp->r_statelock);
2339 
2340 		if (vpm_enable) {
2341 			/*
2342 			 * Copy data. If new pages are created, part of
2343 			 * the page that is not written will be initizliazed
2344 			 * with zeros.
2345 			 */
2346 			error = vpm_data_copy(vp, offset, n, uio,
2347 			    !pagecreate, NULL, 0, S_WRITE);
2348 		} else {
2349 			error = uiomove(base, n, UIO_WRITE, uio);
2350 		}
2351 
2352 		/*
2353 		 * r_size is the maximum number of
2354 		 * bytes known to be in the file.
2355 		 * Make sure it is at least as high as the
2356 		 * first unwritten byte pointed to by uio_loffset.
2357 		 */
2358 		mutex_enter(&rp->r_statelock);
2359 		if (rp->r_size < uio->uio_loffset)
2360 			rp->r_size = uio->uio_loffset;
2361 		rp->r_flags &= ~R4MODINPROGRESS;
2362 		rp->r_flags |= R4DIRTY;
2363 		mutex_exit(&rp->r_statelock);
2364 
2365 		/* n = # of bytes written */
2366 		n = (int)(uio->uio_loffset - offset);
2367 
2368 		if (!vpm_enable) {
2369 			base += n;
2370 		}
2371 
2372 		tcount -= n;
2373 		/*
2374 		 * If we created pages w/o initializing them completely,
2375 		 * we need to zero the part that wasn't set up.
2376 		 * This happens on a most EOF write cases and if
2377 		 * we had some sort of error during the uiomove.
2378 		 */
2379 		if (!vpm_enable && pagecreate) {
2380 			if ((uio->uio_loffset & PAGEOFFSET) || n == 0)
2381 				(void) kzero(base, PAGESIZE - n);
2382 
2383 			if (pgcreated) {
2384 				/*
2385 				 * Caller is responsible for this page,
2386 				 * it was not created in this loop.
2387 				 */
2388 				pgcreated = 0;
2389 			} else {
2390 				/*
2391 				 * For bug 1094402: segmap_pagecreate locks
2392 				 * page. Unlock it. This also unlocks the
2393 				 * pages allocated by page_create_va() in
2394 				 * segmap_pagecreate().
2395 				 */
2396 				sm_error = segmap_fault(kas.a_hat, segkmap,
2397 				    saved_base, saved_n,
2398 				    F_SOFTUNLOCK, S_WRITE);
2399 				if (error == 0)
2400 					error = sm_error;
2401 			}
2402 		}
2403 	} while (tcount > 0 && error == 0);
2404 
2405 	return (error);
2406 }
2407 
2408 int
nfs4_putpages(vnode_t * vp,u_offset_t off,size_t len,int flags,cred_t * cr)2409 nfs4_putpages(vnode_t *vp, u_offset_t off, size_t len, int flags, cred_t *cr)
2410 {
2411 	rnode4_t *rp;
2412 	page_t *pp;
2413 	u_offset_t eoff;
2414 	u_offset_t io_off;
2415 	size_t io_len;
2416 	int error;
2417 	int rdirty;
2418 	int err;
2419 
2420 	rp = VTOR4(vp);
2421 	ASSERT(rp->r_count > 0);
2422 
2423 	if (!nfs4_has_pages(vp))
2424 		return (0);
2425 
2426 	ASSERT(vp->v_type != VCHR);
2427 
2428 	/*
2429 	 * If R4OUTOFSPACE is set, then all writes turn into B_INVAL
2430 	 * writes.  B_FORCE is set to force the VM system to actually
2431 	 * invalidate the pages, even if the i/o failed.  The pages
2432 	 * need to get invalidated because they can't be written out
2433 	 * because there isn't any space left on either the server's
2434 	 * file system or in the user's disk quota.  The B_FREE bit
2435 	 * is cleared to avoid confusion as to whether this is a
2436 	 * request to place the page on the freelist or to destroy
2437 	 * it.
2438 	 */
2439 	if ((rp->r_flags & R4OUTOFSPACE) ||
2440 	    (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED))
2441 		flags = (flags & ~B_FREE) | B_INVAL | B_FORCE;
2442 
2443 	if (len == 0) {
2444 		/*
2445 		 * If doing a full file synchronous operation, then clear
2446 		 * the R4DIRTY bit.  If a page gets dirtied while the flush
2447 		 * is happening, then R4DIRTY will get set again.  The
2448 		 * R4DIRTY bit must get cleared before the flush so that
2449 		 * we don't lose this information.
2450 		 *
2451 		 * If there are no full file async write operations
2452 		 * pending and RDIRTY bit is set, clear it.
2453 		 */
2454 		if (off == (u_offset_t)0 &&
2455 		    !(flags & B_ASYNC) &&
2456 		    (rp->r_flags & R4DIRTY)) {
2457 			mutex_enter(&rp->r_statelock);
2458 			rdirty = (rp->r_flags & R4DIRTY);
2459 			rp->r_flags &= ~R4DIRTY;
2460 			mutex_exit(&rp->r_statelock);
2461 		} else if (flags & B_ASYNC && off == (u_offset_t)0) {
2462 			mutex_enter(&rp->r_statelock);
2463 			if (rp->r_flags & R4DIRTY && rp->r_awcount == 0) {
2464 				rdirty = (rp->r_flags & R4DIRTY);
2465 				rp->r_flags &= ~R4DIRTY;
2466 			}
2467 			mutex_exit(&rp->r_statelock);
2468 		} else
2469 			rdirty = 0;
2470 
2471 		/*
2472 		 * Search the entire vp list for pages >= off, and flush
2473 		 * the dirty pages.
2474 		 */
2475 		error = pvn_vplist_dirty(vp, off, rp->r_putapage,
2476 		    flags, cr);
2477 
2478 		/*
2479 		 * If an error occurred and the file was marked as dirty
2480 		 * before and we aren't forcibly invalidating pages, then
2481 		 * reset the R4DIRTY flag.
2482 		 */
2483 		if (error && rdirty &&
2484 		    (flags & (B_INVAL | B_FORCE)) != (B_INVAL | B_FORCE)) {
2485 			mutex_enter(&rp->r_statelock);
2486 			rp->r_flags |= R4DIRTY;
2487 			mutex_exit(&rp->r_statelock);
2488 		}
2489 	} else {
2490 		/*
2491 		 * Do a range from [off...off + len) looking for pages
2492 		 * to deal with.
2493 		 */
2494 		error = 0;
2495 		io_len = 0;
2496 		eoff = off + len;
2497 		mutex_enter(&rp->r_statelock);
2498 		for (io_off = off; io_off < eoff && io_off < rp->r_size;
2499 		    io_off += io_len) {
2500 			mutex_exit(&rp->r_statelock);
2501 			/*
2502 			 * If we are not invalidating, synchronously
2503 			 * freeing or writing pages use the routine
2504 			 * page_lookup_nowait() to prevent reclaiming
2505 			 * them from the free list.
2506 			 */
2507 			if ((flags & B_INVAL) || !(flags & B_ASYNC)) {
2508 				pp = page_lookup(vp, io_off,
2509 				    (flags & (B_INVAL | B_FREE)) ?
2510 				    SE_EXCL : SE_SHARED);
2511 			} else {
2512 				pp = page_lookup_nowait(vp, io_off,
2513 				    (flags & B_FREE) ? SE_EXCL : SE_SHARED);
2514 			}
2515 
2516 			if (pp == NULL || !pvn_getdirty(pp, flags))
2517 				io_len = PAGESIZE;
2518 			else {
2519 				err = (*rp->r_putapage)(vp, pp, &io_off,
2520 				    &io_len, flags, cr);
2521 				if (!error)
2522 					error = err;
2523 				/*
2524 				 * "io_off" and "io_len" are returned as
2525 				 * the range of pages we actually wrote.
2526 				 * This allows us to skip ahead more quickly
2527 				 * since several pages may've been dealt
2528 				 * with by this iteration of the loop.
2529 				 */
2530 			}
2531 			mutex_enter(&rp->r_statelock);
2532 		}
2533 		mutex_exit(&rp->r_statelock);
2534 	}
2535 
2536 	return (error);
2537 }
2538 
2539 void
nfs4_invalidate_pages(vnode_t * vp,u_offset_t off,cred_t * cr)2540 nfs4_invalidate_pages(vnode_t *vp, u_offset_t off, cred_t *cr)
2541 {
2542 	rnode4_t *rp;
2543 
2544 	rp = VTOR4(vp);
2545 	if (IS_SHADOW(vp, rp))
2546 		vp = RTOV4(rp);
2547 	mutex_enter(&rp->r_statelock);
2548 	while (rp->r_flags & R4TRUNCATE)
2549 		cv_wait(&rp->r_cv, &rp->r_statelock);
2550 	rp->r_flags |= R4TRUNCATE;
2551 	if (off == (u_offset_t)0) {
2552 		rp->r_flags &= ~R4DIRTY;
2553 		if (!(rp->r_flags & R4STALE))
2554 			rp->r_error = 0;
2555 	}
2556 	rp->r_truncaddr = off;
2557 	mutex_exit(&rp->r_statelock);
2558 	(void) pvn_vplist_dirty(vp, off, rp->r_putapage,
2559 	    B_INVAL | B_TRUNC, cr);
2560 	mutex_enter(&rp->r_statelock);
2561 	rp->r_flags &= ~R4TRUNCATE;
2562 	cv_broadcast(&rp->r_cv);
2563 	mutex_exit(&rp->r_statelock);
2564 }
2565 
2566 static int
nfs4_mnt_kstat_update(kstat_t * ksp,int rw)2567 nfs4_mnt_kstat_update(kstat_t *ksp, int rw)
2568 {
2569 	mntinfo4_t *mi;
2570 	struct mntinfo_kstat *mik;
2571 	vfs_t *vfsp;
2572 
2573 	/* this is a read-only kstat. Bail out on a write */
2574 	if (rw == KSTAT_WRITE)
2575 		return (EACCES);
2576 
2577 
2578 	/*
2579 	 * We don't want to wait here as kstat_chain_lock could be held by
2580 	 * dounmount(). dounmount() takes vfs_reflock before the chain lock
2581 	 * and thus could lead to a deadlock.
2582 	 */
2583 	vfsp = (struct vfs *)ksp->ks_private;
2584 
2585 	mi = VFTOMI4(vfsp);
2586 	mik = (struct mntinfo_kstat *)ksp->ks_data;
2587 
2588 	(void) strcpy(mik->mik_proto, mi->mi_curr_serv->sv_knconf->knc_proto);
2589 
2590 	mik->mik_vers = (uint32_t)mi->mi_vers;
2591 	mik->mik_flags = mi->mi_flags;
2592 	/*
2593 	 * The sv_secdata holds the flavor the client specifies.
2594 	 * If the client uses default and a security negotiation
2595 	 * occurs, sv_currsec will point to the current flavor
2596 	 * selected from the server flavor list.
2597 	 * sv_currsec is NULL if no security negotiation takes place.
2598 	 */
2599 	mik->mik_secmod = mi->mi_curr_serv->sv_currsec ?
2600 	    mi->mi_curr_serv->sv_currsec->secmod :
2601 	    mi->mi_curr_serv->sv_secdata->secmod;
2602 	mik->mik_curread = (uint32_t)mi->mi_curread;
2603 	mik->mik_curwrite = (uint32_t)mi->mi_curwrite;
2604 	mik->mik_retrans = mi->mi_retrans;
2605 	mik->mik_timeo = mi->mi_timeo;
2606 	mik->mik_acregmin = HR2SEC(mi->mi_acregmin);
2607 	mik->mik_acregmax = HR2SEC(mi->mi_acregmax);
2608 	mik->mik_acdirmin = HR2SEC(mi->mi_acdirmin);
2609 	mik->mik_acdirmax = HR2SEC(mi->mi_acdirmax);
2610 	mik->mik_noresponse = (uint32_t)mi->mi_noresponse;
2611 	mik->mik_failover = (uint32_t)mi->mi_failover;
2612 	mik->mik_remap = (uint32_t)mi->mi_remap;
2613 
2614 	(void) strcpy(mik->mik_curserver, mi->mi_curr_serv->sv_hostname);
2615 
2616 	return (0);
2617 }
2618 
2619 void
nfs4_mnt_kstat_init(struct vfs * vfsp)2620 nfs4_mnt_kstat_init(struct vfs *vfsp)
2621 {
2622 	mntinfo4_t *mi = VFTOMI4(vfsp);
2623 
2624 	/*
2625 	 * PSARC 2001/697 Contract Private Interface
2626 	 * All nfs kstats are under SunMC contract
2627 	 * Please refer to the PSARC listed above and contact
2628 	 * SunMC before making any changes!
2629 	 *
2630 	 * Changes must be reviewed by Solaris File Sharing
2631 	 * Changes must be communicated to contract-2001-697@sun.com
2632 	 *
2633 	 */
2634 
2635 	mi->mi_io_kstats = kstat_create_zone("nfs", getminor(vfsp->vfs_dev),
2636 	    NULL, "nfs", KSTAT_TYPE_IO, 1, 0, mi->mi_zone->zone_id);
2637 	if (mi->mi_io_kstats) {
2638 		if (mi->mi_zone->zone_id != GLOBAL_ZONEID)
2639 			kstat_zone_add(mi->mi_io_kstats, GLOBAL_ZONEID);
2640 		mi->mi_io_kstats->ks_lock = &mi->mi_lock;
2641 		kstat_install(mi->mi_io_kstats);
2642 	}
2643 
2644 	if ((mi->mi_ro_kstats = kstat_create_zone("nfs",
2645 	    getminor(vfsp->vfs_dev), "mntinfo", "misc", KSTAT_TYPE_RAW,
2646 	    sizeof (struct mntinfo_kstat), 0, mi->mi_zone->zone_id)) != NULL) {
2647 		if (mi->mi_zone->zone_id != GLOBAL_ZONEID)
2648 			kstat_zone_add(mi->mi_ro_kstats, GLOBAL_ZONEID);
2649 		mi->mi_ro_kstats->ks_update = nfs4_mnt_kstat_update;
2650 		mi->mi_ro_kstats->ks_private = (void *)vfsp;
2651 		kstat_install(mi->mi_ro_kstats);
2652 	}
2653 
2654 	nfs4_mnt_recov_kstat_init(vfsp);
2655 }
2656 
2657 void
nfs4_write_error(vnode_t * vp,int error,cred_t * cr)2658 nfs4_write_error(vnode_t *vp, int error, cred_t *cr)
2659 {
2660 	mntinfo4_t *mi;
2661 	clock_t now = ddi_get_lbolt();
2662 
2663 	mi = VTOMI4(vp);
2664 	/*
2665 	 * In case of forced unmount, do not print any messages
2666 	 * since it can flood the console with error messages.
2667 	 */
2668 	if (mi->mi_vfsp->vfs_flag & VFS_UNMOUNTED)
2669 		return;
2670 
2671 	/*
2672 	 * If the mount point is dead, not recoverable, do not
2673 	 * print error messages that can flood the console.
2674 	 */
2675 	if (mi->mi_flags & MI4_RECOV_FAIL)
2676 		return;
2677 
2678 	/*
2679 	 * No use in flooding the console with ENOSPC
2680 	 * messages from the same file system.
2681 	 */
2682 	if ((error != ENOSPC && error != EDQUOT) ||
2683 	    now - mi->mi_printftime > 0) {
2684 		zoneid_t zoneid = mi->mi_zone->zone_id;
2685 
2686 #ifdef DEBUG
2687 		nfs_perror(error, "NFS%ld write error on host %s: %m.\n",
2688 		    mi->mi_vers, VTOR4(vp)->r_server->sv_hostname, NULL);
2689 #else
2690 		nfs_perror(error, "NFS write error on host %s: %m.\n",
2691 		    VTOR4(vp)->r_server->sv_hostname, NULL);
2692 #endif
2693 		if (error == ENOSPC || error == EDQUOT) {
2694 			zcmn_err(zoneid, CE_CONT,
2695 			    "^File: userid=%d, groupid=%d\n",
2696 			    crgetuid(cr), crgetgid(cr));
2697 			if (crgetuid(curthread->t_cred) != crgetuid(cr) ||
2698 			    crgetgid(curthread->t_cred) != crgetgid(cr)) {
2699 				zcmn_err(zoneid, CE_CONT,
2700 				    "^User: userid=%d, groupid=%d\n",
2701 				    crgetuid(curthread->t_cred),
2702 				    crgetgid(curthread->t_cred));
2703 			}
2704 			mi->mi_printftime = now +
2705 			    nfs_write_error_interval * hz;
2706 		}
2707 		sfh4_printfhandle(VTOR4(vp)->r_fh);
2708 #ifdef DEBUG
2709 		if (error == EACCES) {
2710 			zcmn_err(zoneid, CE_CONT,
2711 			    "nfs_bio: cred is%s kcred\n",
2712 			    cr == kcred ? "" : " not");
2713 		}
2714 #endif
2715 	}
2716 }
2717 
2718 /*
2719  * Return non-zero if the given file can be safely memory mapped.  Locks
2720  * are safe if whole-file (length and offset are both zero).
2721  */
2722 
2723 #define	SAFE_LOCK(flk)	((flk).l_start == 0 && (flk).l_len == 0)
2724 
2725 static int
nfs4_safemap(const vnode_t * vp)2726 nfs4_safemap(const vnode_t *vp)
2727 {
2728 	locklist_t	*llp, *next_llp;
2729 	int		safe = 1;
2730 	rnode4_t	*rp = VTOR4(vp);
2731 
2732 	ASSERT(nfs_rw_lock_held(&rp->r_lkserlock, RW_WRITER));
2733 
2734 	NFS4_DEBUG(nfs4_client_map_debug, (CE_NOTE, "nfs4_safemap: "
2735 	    "vp = %p", (void *)vp));
2736 
2737 	/*
2738 	 * Review all the locks for the vnode, both ones that have been
2739 	 * acquired and ones that are pending.  We assume that
2740 	 * flk_active_locks_for_vp() has merged any locks that can be
2741 	 * merged (so that if a process has the entire file locked, it is
2742 	 * represented as a single lock).
2743 	 *
2744 	 * Note that we can't bail out of the loop if we find a non-safe
2745 	 * lock, because we have to free all the elements in the llp list.
2746 	 * We might be able to speed up this code slightly by not looking
2747 	 * at each lock's l_start and l_len fields once we've found a
2748 	 * non-safe lock.
2749 	 */
2750 
2751 	llp = flk_active_locks_for_vp(vp);
2752 	while (llp) {
2753 		NFS4_DEBUG(nfs4_client_map_debug, (CE_NOTE,
2754 		    "nfs4_safemap: active lock (%" PRId64 ", %" PRId64 ")",
2755 		    llp->ll_flock.l_start, llp->ll_flock.l_len));
2756 		if (!SAFE_LOCK(llp->ll_flock)) {
2757 			safe = 0;
2758 			NFS4_DEBUG(nfs4_client_map_debug, (CE_NOTE,
2759 			    "nfs4_safemap: unsafe active lock (%" PRId64
2760 			    ", %" PRId64 ")", llp->ll_flock.l_start,
2761 			    llp->ll_flock.l_len));
2762 		}
2763 		next_llp = llp->ll_next;
2764 		VN_RELE(llp->ll_vp);
2765 		kmem_free(llp, sizeof (*llp));
2766 		llp = next_llp;
2767 	}
2768 
2769 	NFS4_DEBUG(nfs4_client_map_debug, (CE_NOTE, "nfs4_safemap: %s",
2770 	    safe ? "safe" : "unsafe"));
2771 	return (safe);
2772 }
2773 
2774 /*
2775  * Return whether there is a lost LOCK or LOCKU queued up for the given
2776  * file that would make an mmap request unsafe.  cf. nfs4_safemap().
2777  */
2778 
2779 bool_t
nfs4_map_lost_lock_conflict(vnode_t * vp)2780 nfs4_map_lost_lock_conflict(vnode_t *vp)
2781 {
2782 	bool_t conflict = FALSE;
2783 	nfs4_lost_rqst_t *lrp;
2784 	mntinfo4_t *mi = VTOMI4(vp);
2785 
2786 	mutex_enter(&mi->mi_lock);
2787 	for (lrp = list_head(&mi->mi_lost_state); lrp != NULL;
2788 	    lrp = list_next(&mi->mi_lost_state, lrp)) {
2789 		if (lrp->lr_op != OP_LOCK && lrp->lr_op != OP_LOCKU)
2790 			continue;
2791 		ASSERT(lrp->lr_vp != NULL);
2792 		if (!VOP_CMP(lrp->lr_vp, vp, NULL))
2793 			continue;	/* different file */
2794 		if (!SAFE_LOCK(*lrp->lr_flk)) {
2795 			conflict = TRUE;
2796 			break;
2797 		}
2798 	}
2799 
2800 	mutex_exit(&mi->mi_lock);
2801 	return (conflict);
2802 }
2803 
2804 /*
2805  * nfs_lockcompletion:
2806  *
2807  * If the vnode has a lock that makes it unsafe to cache the file, mark it
2808  * as non cachable (set VNOCACHE bit).
2809  */
2810 
2811 void
nfs4_lockcompletion(vnode_t * vp,int cmd)2812 nfs4_lockcompletion(vnode_t *vp, int cmd)
2813 {
2814 	rnode4_t *rp = VTOR4(vp);
2815 
2816 	ASSERT(nfs_rw_lock_held(&rp->r_lkserlock, RW_WRITER));
2817 	ASSERT(!IS_SHADOW(vp, rp));
2818 
2819 	if (cmd == F_SETLK || cmd == F_SETLKW) {
2820 
2821 		if (!nfs4_safemap(vp)) {
2822 			mutex_enter(&vp->v_lock);
2823 			vp->v_flag |= VNOCACHE;
2824 			mutex_exit(&vp->v_lock);
2825 		} else {
2826 			mutex_enter(&vp->v_lock);
2827 			vp->v_flag &= ~VNOCACHE;
2828 			mutex_exit(&vp->v_lock);
2829 		}
2830 	}
2831 	/*
2832 	 * The cached attributes of the file are stale after acquiring
2833 	 * the lock on the file. They were updated when the file was
2834 	 * opened, but not updated when the lock was acquired. Therefore the
2835 	 * cached attributes are invalidated after the lock is obtained.
2836 	 */
2837 	PURGE_ATTRCACHE4(vp);
2838 }
2839 
2840 /* ARGSUSED */
2841 static void *
nfs4_mi_init(zoneid_t zoneid)2842 nfs4_mi_init(zoneid_t zoneid)
2843 {
2844 	struct mi4_globals *mig;
2845 
2846 	mig = kmem_alloc(sizeof (*mig), KM_SLEEP);
2847 	mutex_init(&mig->mig_lock, NULL, MUTEX_DEFAULT, NULL);
2848 	list_create(&mig->mig_list, sizeof (mntinfo4_t),
2849 	    offsetof(mntinfo4_t, mi_zone_node));
2850 	mig->mig_destructor_called = B_FALSE;
2851 	return (mig);
2852 }
2853 
2854 /*
2855  * Callback routine to tell all NFSv4 mounts in the zone to start tearing down
2856  * state and killing off threads.
2857  */
2858 /* ARGSUSED */
2859 static void
nfs4_mi_shutdown(zoneid_t zoneid,void * data)2860 nfs4_mi_shutdown(zoneid_t zoneid, void *data)
2861 {
2862 	struct mi4_globals *mig = data;
2863 	mntinfo4_t *mi;
2864 	nfs4_server_t *np;
2865 
2866 	NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE,
2867 	    "nfs4_mi_shutdown zone %d\n", zoneid));
2868 	ASSERT(mig != NULL);
2869 	for (;;) {
2870 		mutex_enter(&mig->mig_lock);
2871 		mi = list_head(&mig->mig_list);
2872 		if (mi == NULL) {
2873 			mutex_exit(&mig->mig_lock);
2874 			break;
2875 		}
2876 
2877 		NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE,
2878 		    "nfs4_mi_shutdown stopping vfs %p\n", (void *)mi->mi_vfsp));
2879 		/*
2880 		 * purge the DNLC for this filesystem
2881 		 */
2882 		(void) dnlc_purge_vfsp(mi->mi_vfsp, 0);
2883 		/*
2884 		 * Tell existing async worker threads to exit.
2885 		 */
2886 		mutex_enter(&mi->mi_async_lock);
2887 		mi->mi_max_threads = 0;
2888 		NFS4_WAKEALL_ASYNC_WORKERS(mi->mi_async_work_cv);
2889 		/*
2890 		 * Set the appropriate flags, signal and wait for both the
2891 		 * async manager and the inactive thread to exit when they're
2892 		 * done with their current work.
2893 		 */
2894 		mutex_enter(&mi->mi_lock);
2895 		mi->mi_flags |= (MI4_ASYNC_MGR_STOP|MI4_DEAD);
2896 		mutex_exit(&mi->mi_lock);
2897 		mutex_exit(&mi->mi_async_lock);
2898 		if (mi->mi_manager_thread) {
2899 			nfs4_async_manager_stop(mi->mi_vfsp);
2900 		}
2901 		if (mi->mi_inactive_thread) {
2902 			mutex_enter(&mi->mi_async_lock);
2903 			cv_signal(&mi->mi_inact_req_cv);
2904 			/*
2905 			 * Wait for the inactive thread to exit.
2906 			 */
2907 			while (mi->mi_inactive_thread != NULL) {
2908 				cv_wait(&mi->mi_async_cv, &mi->mi_async_lock);
2909 			}
2910 			mutex_exit(&mi->mi_async_lock);
2911 		}
2912 		/*
2913 		 * Wait for the recovery thread to complete, that is, it will
2914 		 * signal when it is done using the "mi" structure and about
2915 		 * to exit
2916 		 */
2917 		mutex_enter(&mi->mi_lock);
2918 		while (mi->mi_in_recovery > 0)
2919 			cv_wait(&mi->mi_cv_in_recov, &mi->mi_lock);
2920 		mutex_exit(&mi->mi_lock);
2921 		/*
2922 		 * We're done when every mi has been done or the list is empty.
2923 		 * This one is done, remove it from the list.
2924 		 */
2925 		list_remove(&mig->mig_list, mi);
2926 		mutex_exit(&mig->mig_lock);
2927 		zone_rele_ref(&mi->mi_zone_ref, ZONE_REF_NFSV4);
2928 
2929 		/*
2930 		 * Release hold on vfs and mi done to prevent race with zone
2931 		 * shutdown. This releases the hold in nfs4_mi_zonelist_add.
2932 		 */
2933 		VFS_RELE(mi->mi_vfsp);
2934 		MI4_RELE(mi);
2935 	}
2936 	/*
2937 	 * Tell each renew thread in the zone to exit
2938 	 */
2939 	mutex_enter(&nfs4_server_lst_lock);
2940 	for (np = nfs4_server_lst.forw; np != &nfs4_server_lst; np = np->forw) {
2941 		mutex_enter(&np->s_lock);
2942 		if (np->zoneid == zoneid) {
2943 			/*
2944 			 * We add another hold onto the nfs4_server_t
2945 			 * because this will make sure tha the nfs4_server_t
2946 			 * stays around until nfs4_callback_fini_zone destroys
2947 			 * the zone. This way, the renew thread can
2948 			 * unconditionally release its holds on the
2949 			 * nfs4_server_t.
2950 			 */
2951 			np->s_refcnt++;
2952 			nfs4_mark_srv_dead(np);
2953 		}
2954 		mutex_exit(&np->s_lock);
2955 	}
2956 	mutex_exit(&nfs4_server_lst_lock);
2957 }
2958 
2959 static void
nfs4_mi_free_globals(struct mi4_globals * mig)2960 nfs4_mi_free_globals(struct mi4_globals *mig)
2961 {
2962 	list_destroy(&mig->mig_list);	/* makes sure the list is empty */
2963 	mutex_destroy(&mig->mig_lock);
2964 	kmem_free(mig, sizeof (*mig));
2965 }
2966 
2967 /* ARGSUSED */
2968 static void
nfs4_mi_destroy(zoneid_t zoneid,void * data)2969 nfs4_mi_destroy(zoneid_t zoneid, void *data)
2970 {
2971 	struct mi4_globals *mig = data;
2972 
2973 	NFS4_DEBUG(nfs4_client_zone_debug, (CE_NOTE,
2974 	    "nfs4_mi_destroy zone %d\n", zoneid));
2975 	ASSERT(mig != NULL);
2976 	mutex_enter(&mig->mig_lock);
2977 	if (list_head(&mig->mig_list) != NULL) {
2978 		/* Still waiting for VFS_FREEVFS() */
2979 		mig->mig_destructor_called = B_TRUE;
2980 		mutex_exit(&mig->mig_lock);
2981 		return;
2982 	}
2983 	nfs4_mi_free_globals(mig);
2984 }
2985 
2986 /*
2987  * Add an NFS mount to the per-zone list of NFS mounts.
2988  */
2989 void
nfs4_mi_zonelist_add(mntinfo4_t * mi)2990 nfs4_mi_zonelist_add(mntinfo4_t *mi)
2991 {
2992 	struct mi4_globals *mig;
2993 
2994 	mig = zone_getspecific(mi4_list_key, mi->mi_zone);
2995 	mutex_enter(&mig->mig_lock);
2996 	list_insert_head(&mig->mig_list, mi);
2997 	/*
2998 	 * hold added to eliminate race with zone shutdown -this will be
2999 	 * released in mi_shutdown
3000 	 */
3001 	MI4_HOLD(mi);
3002 	VFS_HOLD(mi->mi_vfsp);
3003 	mutex_exit(&mig->mig_lock);
3004 }
3005 
3006 /*
3007  * Remove an NFS mount from the per-zone list of NFS mounts.
3008  */
3009 int
nfs4_mi_zonelist_remove(mntinfo4_t * mi)3010 nfs4_mi_zonelist_remove(mntinfo4_t *mi)
3011 {
3012 	struct mi4_globals *mig;
3013 	int ret = 0;
3014 
3015 	mig = zone_getspecific(mi4_list_key, mi->mi_zone);
3016 	mutex_enter(&mig->mig_lock);
3017 	mutex_enter(&mi->mi_lock);
3018 	/* if this mi is marked dead, then the zone already released it */
3019 	if (!(mi->mi_flags & MI4_DEAD)) {
3020 		list_remove(&mig->mig_list, mi);
3021 		mutex_exit(&mi->mi_lock);
3022 
3023 		/* release the holds put on in zonelist_add(). */
3024 		VFS_RELE(mi->mi_vfsp);
3025 		MI4_RELE(mi);
3026 		ret = 1;
3027 	} else {
3028 		mutex_exit(&mi->mi_lock);
3029 	}
3030 
3031 	/*
3032 	 * We can be called asynchronously by VFS_FREEVFS() after the zone
3033 	 * shutdown/destroy callbacks have executed; if so, clean up the zone's
3034 	 * mi globals.
3035 	 */
3036 	if (list_head(&mig->mig_list) == NULL &&
3037 	    mig->mig_destructor_called == B_TRUE) {
3038 		nfs4_mi_free_globals(mig);
3039 		return (ret);
3040 	}
3041 	mutex_exit(&mig->mig_lock);
3042 	return (ret);
3043 }
3044 
3045 void
nfs_free_mi4(mntinfo4_t * mi)3046 nfs_free_mi4(mntinfo4_t *mi)
3047 {
3048 	nfs4_open_owner_t	*foop;
3049 	nfs4_oo_hash_bucket_t   *bucketp;
3050 	nfs4_debug_msg_t	*msgp;
3051 	int i;
3052 	servinfo4_t		*svp;
3053 
3054 	/*
3055 	 * Code introduced here should be carefully evaluated to make
3056 	 * sure none of the freed resources are accessed either directly
3057 	 * or indirectly after freeing them. For eg: Introducing calls to
3058 	 * NFS4_DEBUG that use mntinfo4_t structure member after freeing
3059 	 * the structure members or other routines calling back into NFS
3060 	 * accessing freed mntinfo4_t structure member.
3061 	 */
3062 	mutex_enter(&mi->mi_lock);
3063 	ASSERT(mi->mi_recovthread == NULL);
3064 	ASSERT(mi->mi_flags & MI4_ASYNC_MGR_STOP);
3065 	mutex_exit(&mi->mi_lock);
3066 	mutex_enter(&mi->mi_async_lock);
3067 	ASSERT(mi->mi_threads[NFS4_ASYNC_QUEUE] == 0 &&
3068 	    mi->mi_threads[NFS4_ASYNC_PGOPS_QUEUE] == 0);
3069 	ASSERT(mi->mi_manager_thread == NULL);
3070 	mutex_exit(&mi->mi_async_lock);
3071 	if (mi->mi_io_kstats) {
3072 		kstat_delete(mi->mi_io_kstats);
3073 		mi->mi_io_kstats = NULL;
3074 	}
3075 	if (mi->mi_ro_kstats) {
3076 		kstat_delete(mi->mi_ro_kstats);
3077 		mi->mi_ro_kstats = NULL;
3078 	}
3079 	if (mi->mi_recov_ksp) {
3080 		kstat_delete(mi->mi_recov_ksp);
3081 		mi->mi_recov_ksp = NULL;
3082 	}
3083 	mutex_enter(&mi->mi_msg_list_lock);
3084 	while (msgp = list_head(&mi->mi_msg_list)) {
3085 		list_remove(&mi->mi_msg_list, msgp);
3086 		nfs4_free_msg(msgp);
3087 	}
3088 	mutex_exit(&mi->mi_msg_list_lock);
3089 	list_destroy(&mi->mi_msg_list);
3090 	if (mi->mi_fname != NULL)
3091 		fn_rele(&mi->mi_fname);
3092 	if (mi->mi_rootfh != NULL)
3093 		sfh4_rele(&mi->mi_rootfh);
3094 	if (mi->mi_srvparentfh != NULL)
3095 		sfh4_rele(&mi->mi_srvparentfh);
3096 	svp = mi->mi_servers;
3097 	sv4_free(svp);
3098 	mutex_destroy(&mi->mi_lock);
3099 	mutex_destroy(&mi->mi_async_lock);
3100 	mutex_destroy(&mi->mi_msg_list_lock);
3101 	mutex_destroy(&mi->mi_rnodes_lock);
3102 	nfs_rw_destroy(&mi->mi_recovlock);
3103 	nfs_rw_destroy(&mi->mi_rename_lock);
3104 	nfs_rw_destroy(&mi->mi_fh_lock);
3105 	cv_destroy(&mi->mi_failover_cv);
3106 	cv_destroy(&mi->mi_async_reqs_cv);
3107 	cv_destroy(&mi->mi_async_work_cv[NFS4_ASYNC_QUEUE]);
3108 	cv_destroy(&mi->mi_async_work_cv[NFS4_ASYNC_PGOPS_QUEUE]);
3109 	cv_destroy(&mi->mi_async_cv);
3110 	cv_destroy(&mi->mi_inact_req_cv);
3111 	/*
3112 	 * Destroy the oo hash lists and mutexes for the cred hash table.
3113 	 */
3114 	for (i = 0; i < NFS4_NUM_OO_BUCKETS; i++) {
3115 		bucketp = &(mi->mi_oo_list[i]);
3116 		/* Destroy any remaining open owners on the list */
3117 		foop = list_head(&bucketp->b_oo_hash_list);
3118 		while (foop != NULL) {
3119 			list_remove(&bucketp->b_oo_hash_list, foop);
3120 			nfs4_destroy_open_owner(foop);
3121 			foop = list_head(&bucketp->b_oo_hash_list);
3122 		}
3123 		list_destroy(&bucketp->b_oo_hash_list);
3124 		mutex_destroy(&bucketp->b_lock);
3125 	}
3126 	/*
3127 	 * Empty and destroy the freed open owner list.
3128 	 */
3129 	foop = list_head(&mi->mi_foo_list);
3130 	while (foop != NULL) {
3131 		list_remove(&mi->mi_foo_list, foop);
3132 		nfs4_destroy_open_owner(foop);
3133 		foop = list_head(&mi->mi_foo_list);
3134 	}
3135 	list_destroy(&mi->mi_foo_list);
3136 	list_destroy(&mi->mi_bseqid_list);
3137 	list_destroy(&mi->mi_lost_state);
3138 	list_destroy(&mi->mi_rnodes);
3139 	avl_destroy(&mi->mi_filehandles);
3140 	kmem_free(mi, sizeof (*mi));
3141 }
3142 void
mi_hold(mntinfo4_t * mi)3143 mi_hold(mntinfo4_t *mi)
3144 {
3145 	atomic_inc_32(&mi->mi_count);
3146 	ASSERT(mi->mi_count != 0);
3147 }
3148 
3149 void
mi_rele(mntinfo4_t * mi)3150 mi_rele(mntinfo4_t *mi)
3151 {
3152 	ASSERT(mi->mi_count != 0);
3153 	if (atomic_dec_32_nv(&mi->mi_count) == 0) {
3154 		nfs_free_mi4(mi);
3155 	}
3156 }
3157 
3158 vnode_t    nfs4_xattr_notsupp_vnode;
3159 
3160 void
nfs4_clnt_init(void)3161 nfs4_clnt_init(void)
3162 {
3163 	nfs4_vnops_init();
3164 	(void) nfs4_rnode_init();
3165 	(void) nfs4_shadow_init();
3166 	(void) nfs4_acache_init();
3167 	(void) nfs4_subr_init();
3168 	nfs4_acl_init();
3169 	nfs_idmap_init();
3170 	nfs4_callback_init();
3171 	nfs4_secinfo_init();
3172 #ifdef	DEBUG
3173 	tsd_create(&nfs4_tsd_key, NULL);
3174 #endif
3175 
3176 	/*
3177 	 * Add a CPR callback so that we can update client
3178 	 * lease after a suspend and resume.
3179 	 */
3180 	cid = callb_add(nfs4_client_cpr_callb, 0, CB_CL_CPR_RPC, "nfs4");
3181 
3182 	zone_key_create(&mi4_list_key, nfs4_mi_init, nfs4_mi_shutdown,
3183 	    nfs4_mi_destroy);
3184 
3185 	/*
3186 	 * Initialize the reference count of the notsupp xattr cache vnode to 1
3187 	 * so that it never goes away (VOP_INACTIVE isn't called on it).
3188 	 */
3189 	vn_reinit(&nfs4_xattr_notsupp_vnode);
3190 }
3191 
3192 void
nfs4_clnt_fini(void)3193 nfs4_clnt_fini(void)
3194 {
3195 	(void) zone_key_delete(mi4_list_key);
3196 	nfs4_vnops_fini();
3197 	(void) nfs4_rnode_fini();
3198 	(void) nfs4_shadow_fini();
3199 	(void) nfs4_acache_fini();
3200 	(void) nfs4_subr_fini();
3201 	nfs_idmap_fini();
3202 	nfs4_callback_fini();
3203 	nfs4_secinfo_fini();
3204 #ifdef	DEBUG
3205 	tsd_destroy(&nfs4_tsd_key);
3206 #endif
3207 	if (cid)
3208 		(void) callb_delete(cid);
3209 }
3210 
3211 /*ARGSUSED*/
3212 static boolean_t
nfs4_client_cpr_callb(void * arg,int code)3213 nfs4_client_cpr_callb(void *arg, int code)
3214 {
3215 	/*
3216 	 * We get called for Suspend and Resume events.
3217 	 * For the suspend case we simply don't care!
3218 	 */
3219 	if (code == CB_CODE_CPR_CHKPT) {
3220 		return (B_TRUE);
3221 	}
3222 
3223 	/*
3224 	 * When we get to here we are in the process of
3225 	 * resuming the system from a previous suspend.
3226 	 */
3227 	nfs4_client_resumed = gethrestime_sec();
3228 	return (B_TRUE);
3229 }
3230 
3231 void
nfs4_renew_lease_thread(nfs4_server_t * sp)3232 nfs4_renew_lease_thread(nfs4_server_t *sp)
3233 {
3234 	int	error = 0;
3235 	time_t	tmp_last_renewal_time, tmp_time, tmp_now_time, kip_secs;
3236 	clock_t	tick_delay = 0;
3237 	clock_t time_left = 0;
3238 	callb_cpr_t cpr_info;
3239 	kmutex_t cpr_lock;
3240 
3241 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3242 	    "nfs4_renew_lease_thread: acting on sp 0x%p", (void*)sp));
3243 	mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL);
3244 	CALLB_CPR_INIT(&cpr_info, &cpr_lock, callb_generic_cpr, "nfsv4Lease");
3245 
3246 	mutex_enter(&sp->s_lock);
3247 	/* sp->s_lease_time is set via a GETATTR */
3248 	sp->last_renewal_time = gethrestime_sec();
3249 	sp->lease_valid = NFS4_LEASE_UNINITIALIZED;
3250 	ASSERT(sp->s_refcnt >= 1);
3251 
3252 	for (;;) {
3253 		if (!sp->state_ref_count ||
3254 		    sp->lease_valid != NFS4_LEASE_VALID) {
3255 
3256 			kip_secs = MAX((sp->s_lease_time >> 1) -
3257 			    (3 * sp->propagation_delay.tv_sec), 1);
3258 
3259 			tick_delay = SEC_TO_TICK(kip_secs);
3260 
3261 			NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3262 			    "nfs4_renew_lease_thread: no renew : thread "
3263 			    "wait %ld secs", kip_secs));
3264 
3265 			NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3266 			    "nfs4_renew_lease_thread: no renew : "
3267 			    "state_ref_count %d, lease_valid %d",
3268 			    sp->state_ref_count, sp->lease_valid));
3269 
3270 			mutex_enter(&cpr_lock);
3271 			CALLB_CPR_SAFE_BEGIN(&cpr_info);
3272 			mutex_exit(&cpr_lock);
3273 			time_left = cv_reltimedwait(&sp->cv_thread_exit,
3274 			    &sp->s_lock, tick_delay, TR_CLOCK_TICK);
3275 			mutex_enter(&cpr_lock);
3276 			CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock);
3277 			mutex_exit(&cpr_lock);
3278 
3279 			NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3280 			    "nfs4_renew_lease_thread: no renew: "
3281 			    "time left %ld", time_left));
3282 
3283 			if (sp->s_thread_exit == NFS4_THREAD_EXIT)
3284 				goto die;
3285 			continue;
3286 		}
3287 
3288 		tmp_last_renewal_time = sp->last_renewal_time;
3289 
3290 		tmp_time = gethrestime_sec() - sp->last_renewal_time +
3291 		    (3 * sp->propagation_delay.tv_sec);
3292 
3293 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3294 		    "nfs4_renew_lease_thread: tmp_time %ld, "
3295 		    "sp->last_renewal_time %ld", tmp_time,
3296 		    sp->last_renewal_time));
3297 
3298 		kip_secs = MAX((sp->s_lease_time >> 1) - tmp_time, 1);
3299 
3300 		tick_delay = SEC_TO_TICK(kip_secs);
3301 
3302 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3303 		    "nfs4_renew_lease_thread: valid lease: sleep for %ld "
3304 		    "secs", kip_secs));
3305 
3306 		mutex_enter(&cpr_lock);
3307 		CALLB_CPR_SAFE_BEGIN(&cpr_info);
3308 		mutex_exit(&cpr_lock);
3309 		time_left = cv_reltimedwait(&sp->cv_thread_exit, &sp->s_lock,
3310 		    tick_delay, TR_CLOCK_TICK);
3311 		mutex_enter(&cpr_lock);
3312 		CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock);
3313 		mutex_exit(&cpr_lock);
3314 
3315 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3316 		    "nfs4_renew_lease_thread: valid lease: time left %ld :"
3317 		    "sp last_renewal_time %ld, nfs4_client_resumed %ld, "
3318 		    "tmp_last_renewal_time %ld", time_left,
3319 		    sp->last_renewal_time, nfs4_client_resumed,
3320 		    tmp_last_renewal_time));
3321 
3322 		if (sp->s_thread_exit == NFS4_THREAD_EXIT)
3323 			goto die;
3324 
3325 		if (tmp_last_renewal_time == sp->last_renewal_time ||
3326 		    (nfs4_client_resumed != 0 &&
3327 		    nfs4_client_resumed > sp->last_renewal_time)) {
3328 			/*
3329 			 * Issue RENEW op since we haven't renewed the lease
3330 			 * since we slept.
3331 			 */
3332 			tmp_now_time = gethrestime_sec();
3333 			error = nfs4renew(sp);
3334 			/*
3335 			 * Need to re-acquire sp's lock, nfs4renew()
3336 			 * relinqueshes it.
3337 			 */
3338 			mutex_enter(&sp->s_lock);
3339 
3340 			/*
3341 			 * See if someone changed s_thread_exit while we gave
3342 			 * up s_lock.
3343 			 */
3344 			if (sp->s_thread_exit == NFS4_THREAD_EXIT)
3345 				goto die;
3346 
3347 			if (!error) {
3348 				/*
3349 				 * check to see if we implicitly renewed while
3350 				 * we waited for a reply for our RENEW call.
3351 				 */
3352 				if (tmp_last_renewal_time ==
3353 				    sp->last_renewal_time) {
3354 					/* no implicit renew came */
3355 					sp->last_renewal_time = tmp_now_time;
3356 				} else {
3357 					NFS4_DEBUG(nfs4_client_lease_debug,
3358 					    (CE_NOTE, "renew_thread: did "
3359 					    "implicit renewal before reply "
3360 					    "from server for RENEW"));
3361 				}
3362 			} else {
3363 				/* figure out error */
3364 				NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3365 				    "renew_thread: nfs4renew returned error"
3366 				    " %d", error));
3367 			}
3368 
3369 		}
3370 	}
3371 
3372 die:
3373 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3374 	    "nfs4_renew_lease_thread: thread exiting"));
3375 
3376 	while (sp->s_otw_call_count != 0) {
3377 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3378 		    "nfs4_renew_lease_thread: waiting for outstanding "
3379 		    "otw calls to finish for sp 0x%p, current "
3380 		    "s_otw_call_count %d", (void *)sp,
3381 		    sp->s_otw_call_count));
3382 		mutex_enter(&cpr_lock);
3383 		CALLB_CPR_SAFE_BEGIN(&cpr_info);
3384 		mutex_exit(&cpr_lock);
3385 		cv_wait(&sp->s_cv_otw_count, &sp->s_lock);
3386 		mutex_enter(&cpr_lock);
3387 		CALLB_CPR_SAFE_END(&cpr_info, &cpr_lock);
3388 		mutex_exit(&cpr_lock);
3389 	}
3390 	mutex_exit(&sp->s_lock);
3391 
3392 	nfs4_server_rele(sp);		/* free the thread's reference */
3393 	nfs4_server_rele(sp);		/* free the list's reference */
3394 	sp = NULL;
3395 
3396 done:
3397 	mutex_enter(&cpr_lock);
3398 	CALLB_CPR_EXIT(&cpr_info);	/* drops cpr_lock */
3399 	mutex_destroy(&cpr_lock);
3400 
3401 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3402 	    "nfs4_renew_lease_thread: renew thread exit officially"));
3403 
3404 	zthread_exit();
3405 	/* NOT REACHED */
3406 }
3407 
3408 /*
3409  * Send out a RENEW op to the server.
3410  * Assumes sp is locked down.
3411  */
3412 static int
nfs4renew(nfs4_server_t * sp)3413 nfs4renew(nfs4_server_t *sp)
3414 {
3415 	COMPOUND4args_clnt args;
3416 	COMPOUND4res_clnt res;
3417 	nfs_argop4 argop[1];
3418 	int doqueue = 1;
3419 	int rpc_error;
3420 	cred_t *cr;
3421 	mntinfo4_t *mi;
3422 	timespec_t prop_time, after_time;
3423 	int needrecov = FALSE;
3424 	nfs4_recov_state_t recov_state;
3425 	nfs4_error_t e = { 0, NFS4_OK, RPC_SUCCESS };
3426 
3427 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "nfs4renew"));
3428 
3429 	recov_state.rs_flags = 0;
3430 	recov_state.rs_num_retry_despite_err = 0;
3431 
3432 recov_retry:
3433 	mi = sp->mntinfo4_list;
3434 	VFS_HOLD(mi->mi_vfsp);
3435 	mutex_exit(&sp->s_lock);
3436 	ASSERT(mi != NULL);
3437 
3438 	e.error = nfs4_start_op(mi, NULL, NULL, &recov_state);
3439 	if (e.error) {
3440 		VFS_RELE(mi->mi_vfsp);
3441 		return (e.error);
3442 	}
3443 
3444 	/* Check to see if we're dealing with a marked-dead sp */
3445 	mutex_enter(&sp->s_lock);
3446 	if (sp->s_thread_exit == NFS4_THREAD_EXIT) {
3447 		mutex_exit(&sp->s_lock);
3448 		nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
3449 		VFS_RELE(mi->mi_vfsp);
3450 		return (0);
3451 	}
3452 
3453 	/* Make sure mi hasn't changed on us */
3454 	if (mi != sp->mntinfo4_list) {
3455 		/* Must drop sp's lock to avoid a recursive mutex enter */
3456 		mutex_exit(&sp->s_lock);
3457 		nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
3458 		VFS_RELE(mi->mi_vfsp);
3459 		mutex_enter(&sp->s_lock);
3460 		goto recov_retry;
3461 	}
3462 	mutex_exit(&sp->s_lock);
3463 
3464 	args.ctag = TAG_RENEW;
3465 
3466 	args.array_len = 1;
3467 	args.array = argop;
3468 
3469 	argop[0].argop = OP_RENEW;
3470 
3471 	mutex_enter(&sp->s_lock);
3472 	argop[0].nfs_argop4_u.oprenew.clientid = sp->clientid;
3473 	cr = sp->s_cred;
3474 	crhold(cr);
3475 	mutex_exit(&sp->s_lock);
3476 
3477 	ASSERT(cr != NULL);
3478 
3479 	/* used to figure out RTT for sp */
3480 	gethrestime(&prop_time);
3481 
3482 	NFS4_DEBUG(nfs4_client_call_debug, (CE_NOTE,
3483 	    "nfs4renew: %s call, sp 0x%p", needrecov ? "recov" : "first",
3484 	    (void*)sp));
3485 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "before: %ld s %ld ns ",
3486 	    prop_time.tv_sec, prop_time.tv_nsec));
3487 
3488 	DTRACE_PROBE2(nfs4__renew__start, nfs4_server_t *, sp,
3489 	    mntinfo4_t *, mi);
3490 
3491 	rfs4call(mi, &args, &res, cr, &doqueue, 0, &e);
3492 	crfree(cr);
3493 
3494 	DTRACE_PROBE2(nfs4__renew__end, nfs4_server_t *, sp,
3495 	    mntinfo4_t *, mi);
3496 
3497 	gethrestime(&after_time);
3498 
3499 	mutex_enter(&sp->s_lock);
3500 	sp->propagation_delay.tv_sec =
3501 	    MAX(1, after_time.tv_sec - prop_time.tv_sec);
3502 	mutex_exit(&sp->s_lock);
3503 
3504 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE, "after : %ld s %ld ns ",
3505 	    after_time.tv_sec, after_time.tv_nsec));
3506 
3507 	if (e.error == 0 && res.status == NFS4ERR_CB_PATH_DOWN) {
3508 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3509 		nfs4_delegreturn_all(sp);
3510 		nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
3511 		VFS_RELE(mi->mi_vfsp);
3512 		/*
3513 		 * If the server returns CB_PATH_DOWN, it has renewed
3514 		 * the lease and informed us that the callback path is
3515 		 * down.  Since the lease is renewed, just return 0 and
3516 		 * let the renew thread proceed as normal.
3517 		 */
3518 		return (0);
3519 	}
3520 
3521 	needrecov = nfs4_needs_recovery(&e, FALSE, mi->mi_vfsp);
3522 	if (!needrecov && e.error) {
3523 		nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
3524 		VFS_RELE(mi->mi_vfsp);
3525 		return (e.error);
3526 	}
3527 
3528 	rpc_error = e.error;
3529 
3530 	if (needrecov) {
3531 		NFS4_DEBUG(nfs4_client_recov_debug, (CE_NOTE,
3532 		    "nfs4renew: initiating recovery\n"));
3533 
3534 		if (nfs4_start_recovery(&e, mi, NULL, NULL, NULL, NULL,
3535 		    OP_RENEW, NULL, NULL, NULL) == FALSE) {
3536 			nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
3537 			VFS_RELE(mi->mi_vfsp);
3538 			if (!e.error)
3539 				xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3540 			mutex_enter(&sp->s_lock);
3541 			goto recov_retry;
3542 		}
3543 		/* fall through for res.status case */
3544 	}
3545 
3546 	if (res.status) {
3547 		if (res.status == NFS4ERR_LEASE_MOVED) {
3548 			/*EMPTY*/
3549 			/*
3550 			 * XXX need to try every mntinfo4 in sp->mntinfo4_list
3551 			 * to renew the lease on that server
3552 			 */
3553 		}
3554 		e.error = geterrno4(res.status);
3555 	}
3556 
3557 	if (!rpc_error)
3558 		xdr_free(xdr_COMPOUND4res_clnt, (caddr_t)&res);
3559 
3560 	nfs4_end_op(mi, NULL, NULL, &recov_state, needrecov);
3561 
3562 	VFS_RELE(mi->mi_vfsp);
3563 
3564 	return (e.error);
3565 }
3566 
3567 void
nfs4_inc_state_ref_count(mntinfo4_t * mi)3568 nfs4_inc_state_ref_count(mntinfo4_t *mi)
3569 {
3570 	nfs4_server_t	*sp;
3571 
3572 	/* this locks down sp if it is found */
3573 	sp = find_nfs4_server(mi);
3574 
3575 	if (sp != NULL) {
3576 		nfs4_inc_state_ref_count_nolock(sp, mi);
3577 		mutex_exit(&sp->s_lock);
3578 		nfs4_server_rele(sp);
3579 	}
3580 }
3581 
3582 /*
3583  * Bump the number of OPEN files (ie: those with state) so we know if this
3584  * nfs4_server has any state to maintain a lease for or not.
3585  *
3586  * Also, marks the nfs4_server's lease valid if it hasn't been done so already.
3587  */
3588 void
nfs4_inc_state_ref_count_nolock(nfs4_server_t * sp,mntinfo4_t * mi)3589 nfs4_inc_state_ref_count_nolock(nfs4_server_t *sp, mntinfo4_t *mi)
3590 {
3591 	ASSERT(mutex_owned(&sp->s_lock));
3592 
3593 	sp->state_ref_count++;
3594 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3595 	    "nfs4_inc_state_ref_count: state_ref_count now %d",
3596 	    sp->state_ref_count));
3597 
3598 	if (sp->lease_valid == NFS4_LEASE_UNINITIALIZED)
3599 		sp->lease_valid = NFS4_LEASE_VALID;
3600 
3601 	/*
3602 	 * If this call caused the lease to be marked valid and/or
3603 	 * took the state_ref_count from 0 to 1, then start the time
3604 	 * on lease renewal.
3605 	 */
3606 	if (sp->lease_valid == NFS4_LEASE_VALID && sp->state_ref_count == 1)
3607 		sp->last_renewal_time = gethrestime_sec();
3608 
3609 	/* update the number of open files for mi */
3610 	mi->mi_open_files++;
3611 }
3612 
3613 void
nfs4_dec_state_ref_count(mntinfo4_t * mi)3614 nfs4_dec_state_ref_count(mntinfo4_t *mi)
3615 {
3616 	nfs4_server_t	*sp;
3617 
3618 	/* this locks down sp if it is found */
3619 	sp = find_nfs4_server_all(mi, 1);
3620 
3621 	if (sp != NULL) {
3622 		nfs4_dec_state_ref_count_nolock(sp, mi);
3623 		mutex_exit(&sp->s_lock);
3624 		nfs4_server_rele(sp);
3625 	}
3626 }
3627 
3628 /*
3629  * Decrement the number of OPEN files (ie: those with state) so we know if
3630  * this nfs4_server has any state to maintain a lease for or not.
3631  */
3632 void
nfs4_dec_state_ref_count_nolock(nfs4_server_t * sp,mntinfo4_t * mi)3633 nfs4_dec_state_ref_count_nolock(nfs4_server_t *sp, mntinfo4_t *mi)
3634 {
3635 	ASSERT(mutex_owned(&sp->s_lock));
3636 	ASSERT(sp->state_ref_count != 0);
3637 	sp->state_ref_count--;
3638 
3639 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3640 	    "nfs4_dec_state_ref_count: state ref count now %d",
3641 	    sp->state_ref_count));
3642 
3643 	mi->mi_open_files--;
3644 	NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3645 	    "nfs4_dec_state_ref_count: mi open files %d, v4 flags 0x%x",
3646 	    mi->mi_open_files, mi->mi_flags));
3647 
3648 	/* We don't have to hold the mi_lock to test mi_flags */
3649 	if (mi->mi_open_files == 0 &&
3650 	    (mi->mi_flags & MI4_REMOVE_ON_LAST_CLOSE)) {
3651 		NFS4_DEBUG(nfs4_client_lease_debug, (CE_NOTE,
3652 		    "nfs4_dec_state_ref_count: remove mntinfo4 %p since "
3653 		    "we have closed the last open file", (void*)mi));
3654 		nfs4_remove_mi_from_server(mi, sp);
3655 	}
3656 }
3657 
3658 bool_t
inlease(nfs4_server_t * sp)3659 inlease(nfs4_server_t *sp)
3660 {
3661 	bool_t result;
3662 
3663 	ASSERT(mutex_owned(&sp->s_lock));
3664 
3665 	if (sp->lease_valid == NFS4_LEASE_VALID &&
3666 	    gethrestime_sec() < sp->last_renewal_time + sp->s_lease_time)
3667 		result = TRUE;
3668 	else
3669 		result = FALSE;
3670 
3671 	return (result);
3672 }
3673 
3674 
3675 /*
3676  * Return non-zero if the given nfs4_server_t is going through recovery.
3677  */
3678 
3679 int
nfs4_server_in_recovery(nfs4_server_t * sp)3680 nfs4_server_in_recovery(nfs4_server_t *sp)
3681 {
3682 	return (nfs_rw_lock_held(&sp->s_recovlock, RW_WRITER));
3683 }
3684 
3685 /*
3686  * Compare two shared filehandle objects.  Returns -1, 0, or +1, if the
3687  * first is less than, equal to, or greater than the second.
3688  */
3689 
3690 int
sfh4cmp(const void * p1,const void * p2)3691 sfh4cmp(const void *p1, const void *p2)
3692 {
3693 	const nfs4_sharedfh_t *sfh1 = (const nfs4_sharedfh_t *)p1;
3694 	const nfs4_sharedfh_t *sfh2 = (const nfs4_sharedfh_t *)p2;
3695 
3696 	return (nfs4cmpfh(&sfh1->sfh_fh, &sfh2->sfh_fh));
3697 }
3698 
3699 /*
3700  * Create a table for shared filehandle objects.
3701  */
3702 
3703 void
sfh4_createtab(avl_tree_t * tab)3704 sfh4_createtab(avl_tree_t *tab)
3705 {
3706 	avl_create(tab, sfh4cmp, sizeof (nfs4_sharedfh_t),
3707 	    offsetof(nfs4_sharedfh_t, sfh_tree));
3708 }
3709 
3710 /*
3711  * Return a shared filehandle object for the given filehandle.  The caller
3712  * is responsible for eventually calling sfh4_rele().
3713  */
3714 
3715 nfs4_sharedfh_t *
sfh4_put(const nfs_fh4 * fh,mntinfo4_t * mi,nfs4_sharedfh_t * key)3716 sfh4_put(const nfs_fh4 *fh, mntinfo4_t *mi, nfs4_sharedfh_t *key)
3717 {
3718 	nfs4_sharedfh_t *sfh, *nsfh;
3719 	avl_index_t where;
3720 	nfs4_sharedfh_t skey;
3721 
3722 	if (!key) {
3723 		skey.sfh_fh = *fh;
3724 		key = &skey;
3725 	}
3726 
3727 	nsfh = kmem_alloc(sizeof (nfs4_sharedfh_t), KM_SLEEP);
3728 	nsfh->sfh_fh.nfs_fh4_len = fh->nfs_fh4_len;
3729 	/*
3730 	 * We allocate the largest possible filehandle size because it's
3731 	 * not that big, and it saves us from possibly having to resize the
3732 	 * buffer later.
3733 	 */
3734 	nsfh->sfh_fh.nfs_fh4_val = kmem_alloc(NFS4_FHSIZE, KM_SLEEP);
3735 	bcopy(fh->nfs_fh4_val, nsfh->sfh_fh.nfs_fh4_val, fh->nfs_fh4_len);
3736 	mutex_init(&nsfh->sfh_lock, NULL, MUTEX_DEFAULT, NULL);
3737 	nsfh->sfh_refcnt = 1;
3738 	nsfh->sfh_flags = SFH4_IN_TREE;
3739 	nsfh->sfh_mi = mi;
3740 	NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE, "sfh4_get: new object (%p)",
3741 	    (void *)nsfh));
3742 
3743 	(void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_WRITER, 0);
3744 	sfh = avl_find(&mi->mi_filehandles, key, &where);
3745 	if (sfh != NULL) {
3746 		mutex_enter(&sfh->sfh_lock);
3747 		sfh->sfh_refcnt++;
3748 		mutex_exit(&sfh->sfh_lock);
3749 		nfs_rw_exit(&mi->mi_fh_lock);
3750 		/* free our speculative allocs */
3751 		kmem_free(nsfh->sfh_fh.nfs_fh4_val, NFS4_FHSIZE);
3752 		kmem_free(nsfh, sizeof (nfs4_sharedfh_t));
3753 		return (sfh);
3754 	}
3755 
3756 	avl_insert(&mi->mi_filehandles, nsfh, where);
3757 	nfs_rw_exit(&mi->mi_fh_lock);
3758 
3759 	return (nsfh);
3760 }
3761 
3762 /*
3763  * Return a shared filehandle object for the given filehandle.  The caller
3764  * is responsible for eventually calling sfh4_rele().
3765  */
3766 
3767 nfs4_sharedfh_t *
sfh4_get(const nfs_fh4 * fh,mntinfo4_t * mi)3768 sfh4_get(const nfs_fh4 *fh, mntinfo4_t *mi)
3769 {
3770 	nfs4_sharedfh_t *sfh;
3771 	nfs4_sharedfh_t key;
3772 
3773 	ASSERT(fh->nfs_fh4_len <= NFS4_FHSIZE);
3774 
3775 #ifdef DEBUG
3776 	if (nfs4_sharedfh_debug) {
3777 		nfs4_fhandle_t fhandle;
3778 
3779 		fhandle.fh_len = fh->nfs_fh4_len;
3780 		bcopy(fh->nfs_fh4_val, fhandle.fh_buf, fhandle.fh_len);
3781 		zcmn_err(mi->mi_zone->zone_id, CE_NOTE, "sfh4_get:");
3782 		nfs4_printfhandle(&fhandle);
3783 	}
3784 #endif
3785 
3786 	/*
3787 	 * If there's already an object for the given filehandle, bump the
3788 	 * reference count and return it.  Otherwise, create a new object
3789 	 * and add it to the AVL tree.
3790 	 */
3791 
3792 	key.sfh_fh = *fh;
3793 
3794 	(void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_READER, 0);
3795 	sfh = avl_find(&mi->mi_filehandles, &key, NULL);
3796 	if (sfh != NULL) {
3797 		mutex_enter(&sfh->sfh_lock);
3798 		sfh->sfh_refcnt++;
3799 		NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE,
3800 		    "sfh4_get: found existing %p, new refcnt=%d",
3801 		    (void *)sfh, sfh->sfh_refcnt));
3802 		mutex_exit(&sfh->sfh_lock);
3803 		nfs_rw_exit(&mi->mi_fh_lock);
3804 		return (sfh);
3805 	}
3806 	nfs_rw_exit(&mi->mi_fh_lock);
3807 
3808 	return (sfh4_put(fh, mi, &key));
3809 }
3810 
3811 /*
3812  * Get a reference to the given shared filehandle object.
3813  */
3814 
3815 void
sfh4_hold(nfs4_sharedfh_t * sfh)3816 sfh4_hold(nfs4_sharedfh_t *sfh)
3817 {
3818 	ASSERT(sfh->sfh_refcnt > 0);
3819 
3820 	mutex_enter(&sfh->sfh_lock);
3821 	sfh->sfh_refcnt++;
3822 	NFS4_DEBUG(nfs4_sharedfh_debug,
3823 	    (CE_NOTE, "sfh4_hold %p, new refcnt=%d",
3824 	    (void *)sfh, sfh->sfh_refcnt));
3825 	mutex_exit(&sfh->sfh_lock);
3826 }
3827 
3828 /*
3829  * Release a reference to the given shared filehandle object and null out
3830  * the given pointer.
3831  */
3832 
3833 void
sfh4_rele(nfs4_sharedfh_t ** sfhpp)3834 sfh4_rele(nfs4_sharedfh_t **sfhpp)
3835 {
3836 	mntinfo4_t *mi;
3837 	nfs4_sharedfh_t *sfh = *sfhpp;
3838 
3839 	ASSERT(sfh->sfh_refcnt > 0);
3840 
3841 	mutex_enter(&sfh->sfh_lock);
3842 	if (sfh->sfh_refcnt > 1) {
3843 		sfh->sfh_refcnt--;
3844 		NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE,
3845 		    "sfh4_rele %p, new refcnt=%d",
3846 		    (void *)sfh, sfh->sfh_refcnt));
3847 		mutex_exit(&sfh->sfh_lock);
3848 		goto finish;
3849 	}
3850 	mutex_exit(&sfh->sfh_lock);
3851 
3852 	/*
3853 	 * Possibly the last reference, so get the lock for the table in
3854 	 * case it's time to remove the object from the table.
3855 	 */
3856 	mi = sfh->sfh_mi;
3857 	(void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_WRITER, 0);
3858 	mutex_enter(&sfh->sfh_lock);
3859 	sfh->sfh_refcnt--;
3860 	if (sfh->sfh_refcnt > 0) {
3861 		NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE,
3862 		    "sfh4_rele %p, new refcnt=%d",
3863 		    (void *)sfh, sfh->sfh_refcnt));
3864 		mutex_exit(&sfh->sfh_lock);
3865 		nfs_rw_exit(&mi->mi_fh_lock);
3866 		goto finish;
3867 	}
3868 
3869 	NFS4_DEBUG(nfs4_sharedfh_debug, (CE_NOTE,
3870 	    "sfh4_rele %p, last ref", (void *)sfh));
3871 	if (sfh->sfh_flags & SFH4_IN_TREE) {
3872 		avl_remove(&mi->mi_filehandles, sfh);
3873 		sfh->sfh_flags &= ~SFH4_IN_TREE;
3874 	}
3875 	mutex_exit(&sfh->sfh_lock);
3876 	nfs_rw_exit(&mi->mi_fh_lock);
3877 	mutex_destroy(&sfh->sfh_lock);
3878 	kmem_free(sfh->sfh_fh.nfs_fh4_val, NFS4_FHSIZE);
3879 	kmem_free(sfh, sizeof (nfs4_sharedfh_t));
3880 
3881 finish:
3882 	*sfhpp = NULL;
3883 }
3884 
3885 /*
3886  * Update the filehandle for the given shared filehandle object.
3887  */
3888 
3889 int nfs4_warn_dupfh = 0;	/* if set, always warn about dup fhs below */
3890 
3891 void
sfh4_update(nfs4_sharedfh_t * sfh,const nfs_fh4 * newfh)3892 sfh4_update(nfs4_sharedfh_t *sfh, const nfs_fh4 *newfh)
3893 {
3894 	mntinfo4_t *mi = sfh->sfh_mi;
3895 	nfs4_sharedfh_t *dupsfh;
3896 	avl_index_t where;
3897 	nfs4_sharedfh_t key;
3898 
3899 #ifdef DEBUG
3900 	mutex_enter(&sfh->sfh_lock);
3901 	ASSERT(sfh->sfh_refcnt > 0);
3902 	mutex_exit(&sfh->sfh_lock);
3903 #endif
3904 	ASSERT(newfh->nfs_fh4_len <= NFS4_FHSIZE);
3905 
3906 	/*
3907 	 * The basic plan is to remove the shared filehandle object from
3908 	 * the table, update it to have the new filehandle, then reinsert
3909 	 * it.
3910 	 */
3911 
3912 	(void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_WRITER, 0);
3913 	mutex_enter(&sfh->sfh_lock);
3914 	if (sfh->sfh_flags & SFH4_IN_TREE) {
3915 		avl_remove(&mi->mi_filehandles, sfh);
3916 		sfh->sfh_flags &= ~SFH4_IN_TREE;
3917 	}
3918 	mutex_exit(&sfh->sfh_lock);
3919 	sfh->sfh_fh.nfs_fh4_len = newfh->nfs_fh4_len;
3920 	bcopy(newfh->nfs_fh4_val, sfh->sfh_fh.nfs_fh4_val,
3921 	    sfh->sfh_fh.nfs_fh4_len);
3922 
3923 	/*
3924 	 * XXX If there is already a shared filehandle object with the new
3925 	 * filehandle, we're in trouble, because the rnode code assumes
3926 	 * that there is only one shared filehandle object for a given
3927 	 * filehandle.  So issue a warning (for read-write mounts only)
3928 	 * and don't try to re-insert the given object into the table.
3929 	 * Hopefully the given object will quickly go away and everyone
3930 	 * will use the new object.
3931 	 */
3932 	key.sfh_fh = *newfh;
3933 	dupsfh = avl_find(&mi->mi_filehandles, &key, &where);
3934 	if (dupsfh != NULL) {
3935 		if (!(mi->mi_vfsp->vfs_flag & VFS_RDONLY) || nfs4_warn_dupfh) {
3936 			zcmn_err(mi->mi_zone->zone_id, CE_WARN, "sfh4_update: "
3937 			    "duplicate filehandle detected");
3938 			sfh4_printfhandle(dupsfh);
3939 		}
3940 	} else {
3941 		avl_insert(&mi->mi_filehandles, sfh, where);
3942 		mutex_enter(&sfh->sfh_lock);
3943 		sfh->sfh_flags |= SFH4_IN_TREE;
3944 		mutex_exit(&sfh->sfh_lock);
3945 	}
3946 	nfs_rw_exit(&mi->mi_fh_lock);
3947 }
3948 
3949 /*
3950  * Copy out the current filehandle for the given shared filehandle object.
3951  */
3952 
3953 void
sfh4_copyval(const nfs4_sharedfh_t * sfh,nfs4_fhandle_t * fhp)3954 sfh4_copyval(const nfs4_sharedfh_t *sfh, nfs4_fhandle_t *fhp)
3955 {
3956 	mntinfo4_t *mi = sfh->sfh_mi;
3957 
3958 	ASSERT(sfh->sfh_refcnt > 0);
3959 
3960 	(void) nfs_rw_enter_sig(&mi->mi_fh_lock, RW_READER, 0);
3961 	fhp->fh_len = sfh->sfh_fh.nfs_fh4_len;
3962 	ASSERT(fhp->fh_len <= NFS4_FHSIZE);
3963 	bcopy(sfh->sfh_fh.nfs_fh4_val, fhp->fh_buf, fhp->fh_len);
3964 	nfs_rw_exit(&mi->mi_fh_lock);
3965 }
3966 
3967 /*
3968  * Print out the filehandle for the given shared filehandle object.
3969  */
3970 
3971 void
sfh4_printfhandle(const nfs4_sharedfh_t * sfh)3972 sfh4_printfhandle(const nfs4_sharedfh_t *sfh)
3973 {
3974 	nfs4_fhandle_t fhandle;
3975 
3976 	sfh4_copyval(sfh, &fhandle);
3977 	nfs4_printfhandle(&fhandle);
3978 }
3979 
3980 /*
3981  * Compare 2 fnames.  Returns -1 if the first is "less" than the second, 0
3982  * if they're the same, +1 if the first is "greater" than the second.  The
3983  * caller (or whoever's calling the AVL package) is responsible for
3984  * handling locking issues.
3985  */
3986 
3987 static int
fncmp(const void * p1,const void * p2)3988 fncmp(const void *p1, const void *p2)
3989 {
3990 	const nfs4_fname_t *f1 = p1;
3991 	const nfs4_fname_t *f2 = p2;
3992 	int res;
3993 
3994 	res = strcmp(f1->fn_name, f2->fn_name);
3995 	/*
3996 	 * The AVL package wants +/-1, not arbitrary positive or negative
3997 	 * integers.
3998 	 */
3999 	if (res > 0)
4000 		res = 1;
4001 	else if (res < 0)
4002 		res = -1;
4003 	return (res);
4004 }
4005 
4006 /*
4007  * Get or create an fname with the given name, as a child of the given
4008  * fname.  The caller is responsible for eventually releasing the reference
4009  * (fn_rele()).  parent may be NULL.
4010  */
4011 
4012 nfs4_fname_t *
fn_get(nfs4_fname_t * parent,char * name,nfs4_sharedfh_t * sfh)4013 fn_get(nfs4_fname_t *parent, char *name, nfs4_sharedfh_t *sfh)
4014 {
4015 	nfs4_fname_t key;
4016 	nfs4_fname_t *fnp;
4017 	avl_index_t where;
4018 
4019 	key.fn_name = name;
4020 
4021 	/*
4022 	 * If there's already an fname registered with the given name, bump
4023 	 * its reference count and return it.  Otherwise, create a new one
4024 	 * and add it to the parent's AVL tree.
4025 	 *
4026 	 * fname entries we are looking for should match both name
4027 	 * and sfh stored in the fname.
4028 	 */
4029 again:
4030 	if (parent != NULL) {
4031 		mutex_enter(&parent->fn_lock);
4032 		fnp = avl_find(&parent->fn_children, &key, &where);
4033 		if (fnp != NULL) {
4034 			/*
4035 			 * This hold on fnp is released below later,
4036 			 * in case this is not the fnp we want.
4037 			 */
4038 			fn_hold(fnp);
4039 
4040 			if (fnp->fn_sfh == sfh) {
4041 				/*
4042 				 * We have found our entry.
4043 				 * put an hold and return it.
4044 				 */
4045 				mutex_exit(&parent->fn_lock);
4046 				return (fnp);
4047 			}
4048 
4049 			/*
4050 			 * We have found an entry that has a mismatching
4051 			 * fn_sfh. This could be a stale entry due to
4052 			 * server side rename. We will remove this entry
4053 			 * and make sure no such entries exist.
4054 			 */
4055 			mutex_exit(&parent->fn_lock);
4056 			mutex_enter(&fnp->fn_lock);
4057 			if (fnp->fn_parent == parent) {
4058 				/*
4059 				 * Remove ourselves from parent's
4060 				 * fn_children tree.
4061 				 */
4062 				mutex_enter(&parent->fn_lock);
4063 				avl_remove(&parent->fn_children, fnp);
4064 				mutex_exit(&parent->fn_lock);
4065 				fn_rele(&fnp->fn_parent);
4066 			}
4067 			mutex_exit(&fnp->fn_lock);
4068 			fn_rele(&fnp);
4069 			goto again;
4070 		}
4071 	}
4072 
4073 	fnp = kmem_alloc(sizeof (nfs4_fname_t), KM_SLEEP);
4074 	mutex_init(&fnp->fn_lock, NULL, MUTEX_DEFAULT, NULL);
4075 	fnp->fn_parent = parent;
4076 	if (parent != NULL)
4077 		fn_hold(parent);
4078 	fnp->fn_len = strlen(name);
4079 	ASSERT(fnp->fn_len < MAXNAMELEN);
4080 	fnp->fn_name = kmem_alloc(fnp->fn_len + 1, KM_SLEEP);
4081 	(void) strcpy(fnp->fn_name, name);
4082 	fnp->fn_refcnt = 1;
4083 
4084 	/*
4085 	 * This hold on sfh is later released
4086 	 * when we do the final fn_rele() on this fname.
4087 	 */
4088 	sfh4_hold(sfh);
4089 	fnp->fn_sfh = sfh;
4090 
4091 	avl_create(&fnp->fn_children, fncmp, sizeof (nfs4_fname_t),
4092 	    offsetof(nfs4_fname_t, fn_tree));
4093 	NFS4_DEBUG(nfs4_fname_debug, (CE_NOTE,
4094 	    "fn_get %p:%s, a new nfs4_fname_t!",
4095 	    (void *)fnp, fnp->fn_name));
4096 	if (parent != NULL) {
4097 		avl_insert(&parent->fn_children, fnp, where);
4098 		mutex_exit(&parent->fn_lock);
4099 	}
4100 
4101 	return (fnp);
4102 }
4103 
4104 void
fn_hold(nfs4_fname_t * fnp)4105 fn_hold(nfs4_fname_t *fnp)
4106 {
4107 	atomic_inc_32(&fnp->fn_refcnt);
4108 	NFS4_DEBUG(nfs4_fname_debug, (CE_NOTE,
4109 	    "fn_hold %p:%s, new refcnt=%d",
4110 	    (void *)fnp, fnp->fn_name, fnp->fn_refcnt));
4111 }
4112 
4113 /*
4114  * Decrement the reference count of the given fname, and destroy it if its
4115  * reference count goes to zero.  Nulls out the given pointer.
4116  */
4117 
4118 void
fn_rele(nfs4_fname_t ** fnpp)4119 fn_rele(nfs4_fname_t **fnpp)
4120 {
4121 	nfs4_fname_t *parent;
4122 	uint32_t newref;
4123 	nfs4_fname_t *fnp;
4124 
4125 recur:
4126 	fnp = *fnpp;
4127 	*fnpp = NULL;
4128 
4129 	mutex_enter(&fnp->fn_lock);
4130 	parent = fnp->fn_parent;
4131 	if (parent != NULL)
4132 		mutex_enter(&parent->fn_lock);	/* prevent new references */
4133 	newref = atomic_dec_32_nv(&fnp->fn_refcnt);
4134 	if (newref > 0) {
4135 		NFS4_DEBUG(nfs4_fname_debug, (CE_NOTE,
4136 		    "fn_rele %p:%s, new refcnt=%d",
4137 		    (void *)fnp, fnp->fn_name, fnp->fn_refcnt));
4138 		if (parent != NULL)
4139 			mutex_exit(&parent->fn_lock);
4140 		mutex_exit(&fnp->fn_lock);
4141 		return;
4142 	}
4143 
4144 	NFS4_DEBUG(nfs4_fname_debug, (CE_NOTE,
4145 	    "fn_rele %p:%s, last reference, deleting...",
4146 	    (void *)fnp, fnp->fn_name));
4147 	if (parent != NULL) {
4148 		avl_remove(&parent->fn_children, fnp);
4149 		mutex_exit(&parent->fn_lock);
4150 	}
4151 	kmem_free(fnp->fn_name, fnp->fn_len + 1);
4152 	sfh4_rele(&fnp->fn_sfh);
4153 	mutex_destroy(&fnp->fn_lock);
4154 	avl_destroy(&fnp->fn_children);
4155 	kmem_free(fnp, sizeof (nfs4_fname_t));
4156 	/*
4157 	 * Recursivly fn_rele the parent.
4158 	 * Use goto instead of a recursive call to avoid stack overflow.
4159 	 */
4160 	if (parent != NULL) {
4161 		fnpp = &parent;
4162 		goto recur;
4163 	}
4164 }
4165 
4166 /*
4167  * Returns the single component name of the given fname, in a MAXNAMELEN
4168  * string buffer, which the caller is responsible for freeing.  Note that
4169  * the name may become invalid as a result of fn_move().
4170  */
4171 
4172 char *
fn_name(nfs4_fname_t * fnp)4173 fn_name(nfs4_fname_t *fnp)
4174 {
4175 	char *name;
4176 
4177 	ASSERT(fnp->fn_len < MAXNAMELEN);
4178 	name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
4179 	mutex_enter(&fnp->fn_lock);
4180 	(void) strcpy(name, fnp->fn_name);
4181 	mutex_exit(&fnp->fn_lock);
4182 
4183 	return (name);
4184 }
4185 
4186 
4187 /*
4188  * fn_path_realloc
4189  *
4190  * This function, used only by fn_path, constructs
4191  * a new string which looks like "prepend" + "/" + "current".
4192  * by allocating a new string and freeing the old one.
4193  */
4194 static void
fn_path_realloc(char ** curses,char * prepend)4195 fn_path_realloc(char **curses, char *prepend)
4196 {
4197 	int len, curlen = 0;
4198 	char *news;
4199 
4200 	if (*curses == NULL) {
4201 		/*
4202 		 * Prime the pump, allocate just the
4203 		 * space for prepend and return that.
4204 		 */
4205 		len = strlen(prepend) + 1;
4206 		news = kmem_alloc(len, KM_SLEEP);
4207 		(void) strncpy(news, prepend, len);
4208 	} else {
4209 		/*
4210 		 * Allocate the space  for a new string
4211 		 * +1 +1 is for the "/" and the NULL
4212 		 * byte at the end of it all.
4213 		 */
4214 		curlen = strlen(*curses);
4215 		len = curlen + strlen(prepend) + 1 + 1;
4216 		news = kmem_alloc(len, KM_SLEEP);
4217 		(void) strncpy(news, prepend, len);
4218 		(void) strcat(news, "/");
4219 		(void) strcat(news, *curses);
4220 		kmem_free(*curses, curlen + 1);
4221 	}
4222 	*curses = news;
4223 }
4224 
4225 /*
4226  * Returns the path name (starting from the fs root) for the given fname.
4227  * The caller is responsible for freeing.  Note that the path may be or
4228  * become invalid as a result of fn_move().
4229  */
4230 
4231 char *
fn_path(nfs4_fname_t * fnp)4232 fn_path(nfs4_fname_t *fnp)
4233 {
4234 	char *path;
4235 	nfs4_fname_t *nextfnp;
4236 
4237 	if (fnp == NULL)
4238 		return (NULL);
4239 
4240 	path = NULL;
4241 
4242 	/* walk up the tree constructing the pathname.  */
4243 
4244 	fn_hold(fnp);			/* adjust for later rele */
4245 	do {
4246 		mutex_enter(&fnp->fn_lock);
4247 		/*
4248 		 * Add fn_name in front of the current path
4249 		 */
4250 		fn_path_realloc(&path, fnp->fn_name);
4251 		nextfnp = fnp->fn_parent;
4252 		if (nextfnp != NULL)
4253 			fn_hold(nextfnp);
4254 		mutex_exit(&fnp->fn_lock);
4255 		fn_rele(&fnp);
4256 		fnp = nextfnp;
4257 	} while (fnp != NULL);
4258 
4259 	return (path);
4260 }
4261 
4262 /*
4263  * Return a reference to the parent of the given fname, which the caller is
4264  * responsible for eventually releasing.
4265  */
4266 
4267 nfs4_fname_t *
fn_parent(nfs4_fname_t * fnp)4268 fn_parent(nfs4_fname_t *fnp)
4269 {
4270 	nfs4_fname_t *parent;
4271 
4272 	mutex_enter(&fnp->fn_lock);
4273 	parent = fnp->fn_parent;
4274 	if (parent != NULL)
4275 		fn_hold(parent);
4276 	mutex_exit(&fnp->fn_lock);
4277 
4278 	return (parent);
4279 }
4280 
4281 /*
4282  * Update fnp so that its parent is newparent and its name is newname.
4283  */
4284 
4285 void
fn_move(nfs4_fname_t * fnp,nfs4_fname_t * newparent,char * newname)4286 fn_move(nfs4_fname_t *fnp, nfs4_fname_t *newparent, char *newname)
4287 {
4288 	nfs4_fname_t *parent, *tmpfnp;
4289 	ssize_t newlen;
4290 	nfs4_fname_t key;
4291 	avl_index_t where;
4292 
4293 	/*
4294 	 * This assert exists to catch the client trying to rename
4295 	 * a dir to be a child of itself.  This happened at a recent
4296 	 * bakeoff against a 3rd party (broken) server which allowed
4297 	 * the rename to succeed.  If it trips it means that:
4298 	 *	a) the code in nfs4rename that detects this case is broken
4299 	 *	b) the server is broken (since it allowed the bogus rename)
4300 	 *
4301 	 * For non-DEBUG kernels, prepare for a recursive mutex_enter
4302 	 * panic below from:  mutex_enter(&newparent->fn_lock);
4303 	 */
4304 	ASSERT(fnp != newparent);
4305 
4306 	/*
4307 	 * Remove fnp from its current parent, change its name, then add it
4308 	 * to newparent. It might happen that fnp was replaced by another
4309 	 * nfs4_fname_t with the same fn_name in parent->fn_children.
4310 	 * In such case, fnp->fn_parent is NULL and we skip the removal
4311 	 * of fnp from its current parent.
4312 	 */
4313 	mutex_enter(&fnp->fn_lock);
4314 	parent = fnp->fn_parent;
4315 	if (parent != NULL) {
4316 		mutex_enter(&parent->fn_lock);
4317 		avl_remove(&parent->fn_children, fnp);
4318 		mutex_exit(&parent->fn_lock);
4319 		fn_rele(&fnp->fn_parent);
4320 	}
4321 
4322 	newlen = strlen(newname);
4323 	if (newlen != fnp->fn_len) {
4324 		ASSERT(newlen < MAXNAMELEN);
4325 		kmem_free(fnp->fn_name, fnp->fn_len + 1);
4326 		fnp->fn_name = kmem_alloc(newlen + 1, KM_SLEEP);
4327 		fnp->fn_len = newlen;
4328 	}
4329 	(void) strcpy(fnp->fn_name, newname);
4330 
4331 again:
4332 	mutex_enter(&newparent->fn_lock);
4333 	key.fn_name = fnp->fn_name;
4334 	tmpfnp = avl_find(&newparent->fn_children, &key, &where);
4335 	if (tmpfnp != NULL) {
4336 		/*
4337 		 * This could be due to a file that was unlinked while
4338 		 * open, or perhaps the rnode is in the free list.  Remove
4339 		 * it from newparent and let it go away on its own.  The
4340 		 * contorted code is to deal with lock order issues and
4341 		 * race conditions.
4342 		 */
4343 		fn_hold(tmpfnp);
4344 		mutex_exit(&newparent->fn_lock);
4345 		mutex_enter(&tmpfnp->fn_lock);
4346 		if (tmpfnp->fn_parent == newparent) {
4347 			mutex_enter(&newparent->fn_lock);
4348 			avl_remove(&newparent->fn_children, tmpfnp);
4349 			mutex_exit(&newparent->fn_lock);
4350 			fn_rele(&tmpfnp->fn_parent);
4351 		}
4352 		mutex_exit(&tmpfnp->fn_lock);
4353 		fn_rele(&tmpfnp);
4354 		goto again;
4355 	}
4356 	fnp->fn_parent = newparent;
4357 	fn_hold(newparent);
4358 	avl_insert(&newparent->fn_children, fnp, where);
4359 	mutex_exit(&newparent->fn_lock);
4360 	mutex_exit(&fnp->fn_lock);
4361 }
4362 
4363 #ifdef DEBUG
4364 /*
4365  * Return non-zero if the type information makes sense for the given vnode.
4366  * Otherwise panic.
4367  */
4368 int
nfs4_consistent_type(vnode_t * vp)4369 nfs4_consistent_type(vnode_t *vp)
4370 {
4371 	rnode4_t *rp = VTOR4(vp);
4372 
4373 	if (nfs4_vtype_debug && vp->v_type != VNON &&
4374 	    rp->r_attr.va_type != VNON && vp->v_type != rp->r_attr.va_type) {
4375 		cmn_err(CE_PANIC, "vnode %p type mismatch; v_type=%d, "
4376 		    "rnode attr type=%d", (void *)vp, vp->v_type,
4377 		    rp->r_attr.va_type);
4378 	}
4379 
4380 	return (1);
4381 }
4382 #endif /* DEBUG */
4383