xref: /titanic_41/usr/src/uts/common/fs/cachefs/cachefs_resource.c (revision b60d385649ee1c6aa96c1c625333428cf323c393)
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 2007 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 #pragma ident	"%Z%%M%	%I%	%E% SMI"
26 
27 #include <sys/param.h>
28 #include <sys/types.h>
29 #include <sys/systm.h>
30 #include <sys/cred.h>
31 #include <sys/proc.h>
32 #include <sys/user.h>
33 #include <sys/vfs.h>
34 #include <sys/vnode.h>
35 #include <sys/pathname.h>
36 #include <sys/uio.h>
37 #include <sys/tiuser.h>
38 #include <sys/sysmacros.h>
39 #include <sys/kmem.h>
40 #include <sys/kobj.h>
41 #include <sys/mount.h>
42 #include <sys/ioctl.h>
43 #include <sys/statvfs.h>
44 #include <sys/errno.h>
45 #include <sys/debug.h>
46 #include <sys/cmn_err.h>
47 #include <sys/utsname.h>
48 #include <sys/modctl.h>
49 #include <sys/file.h>
50 #include <sys/stat.h>
51 #include <sys/fcntl.h>
52 #include <sys/callb.h>
53 
54 #include <vm/hat.h>
55 #include <vm/as.h>
56 #include <vm/page.h>
57 #include <vm/pvn.h>
58 #include <vm/seg.h>
59 #include <vm/seg_map.h>
60 #include <vm/seg_vn.h>
61 #include <vm/rm.h>
62 #include <sys/fs/cachefs_fs.h>
63 
64 extern time_t time;
65 
66 /* forward references */
67 int cachefs_rl_entry_get(cachefscache_t *, uint_t, rl_entry_t **);
68 void cachefs_garbage_collect_queue(cachefscache_t *cachep);
69 static time_t cachefs_gc_front_atime(cachefscache_t *cachep);
70 static void cachefs_garbage_collect(cachefscache_t *cachep);
71 static void cachefs_packed_pending(cachefscache_t *cachep);
72 
73 
74 #define	RL_HEAD(cachep, type) \
75 	(&(cachep->c_rlinfo.rl_items[CACHEFS_RL_INDEX(type)]))
76 
77 /*
78  * This function moves an RL entry from wherever it currently is to
79  * the back of the requested list.
80  */
81 void
82 cachefs_rlent_moveto(cachefscache_t *cachep,
83     enum cachefs_rl_type type, uint_t entno, size_t blks)
84 {
85 	mutex_enter(&cachep->c_contentslock);
86 	cachefs_cache_dirty(cachep, 0);
87 	cachefs_rlent_moveto_nolock(cachep, type, entno, blks);
88 	mutex_exit(&cachep->c_contentslock);
89 }
90 
91 void
92 cachefs_rlent_moveto_nolock(cachefscache_t *cachep,
93     enum cachefs_rl_type type, uint_t entno, size_t blks)
94 {
95 	rl_entry_t *rl_ent;
96 	uint_t prev, next;
97 	cachefs_rl_listhead_t *lhp;
98 	enum cachefs_rl_type otype;
99 	int error;
100 
101 	ASSERT(entno != 0);
102 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
103 	ASSERT(MUTEX_HELD(&cachep->c_contentslock));
104 	ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
105 
106 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
107 	if (error)
108 		return;
109 	next = rl_ent->rl_fwd_idx;
110 	prev = rl_ent->rl_bkwd_idx;
111 	otype = rl_ent->rl_current;
112 	ASSERT((CACHEFS_RL_START <= otype) && (otype <= CACHEFS_RL_END));
113 	rl_ent->rl_current = CACHEFS_RL_NONE;
114 
115 	if (type == CACHEFS_RL_PACKED_PENDING) {
116 		/* XXX sam: is this the right place to turn this on? */
117 		cachep->c_flags |= CACHE_PACKED_PENDING;
118 	}
119 
120 	/* remove entry from its previous list */
121 
122 	lhp = RL_HEAD(cachep, otype);
123 	if ((lhp->rli_back == 0) || (lhp->rli_front == 0))
124 		ASSERT((lhp->rli_back == 0) && (lhp->rli_front == 0));
125 
126 	if (lhp->rli_back == entno)
127 		lhp->rli_back = next;
128 	if (lhp->rli_front == entno)
129 		lhp->rli_front = prev;
130 	if (prev != 0) {
131 		error = cachefs_rl_entry_get(cachep, prev, &rl_ent);
132 		if (error)
133 			return;
134 		rl_ent->rl_fwd_idx = next;
135 	}
136 	if (next != 0) {
137 		error = cachefs_rl_entry_get(cachep, next, &rl_ent);
138 		if (error)
139 			return;
140 		rl_ent->rl_bkwd_idx = prev;
141 	}
142 	lhp->rli_blkcnt -= blks;
143 	lhp->rli_itemcnt--;
144 
145 	/* add entry to its new list */
146 
147 	lhp = RL_HEAD(cachep, type);
148 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
149 	if (error)
150 		return;
151 	rl_ent->rl_current = type;
152 	rl_ent->rl_bkwd_idx = 0;
153 	rl_ent->rl_fwd_idx = lhp->rli_back;
154 
155 	if (lhp->rli_back != 0) {
156 		ASSERT(lhp->rli_front != 0);
157 		error = cachefs_rl_entry_get(cachep, lhp->rli_back, &rl_ent);
158 		if (error)
159 			return;
160 		rl_ent->rl_bkwd_idx = entno;
161 	} else {
162 		ASSERT(lhp->rli_front == 0);
163 		lhp->rli_front = entno;
164 	}
165 	lhp->rli_back = entno;
166 	lhp->rli_blkcnt += blks;
167 	lhp->rli_itemcnt++;
168 }
169 
170 /*
171  * This function verifies that an rl entry is of the `correct' type.
172  * it's used for debugging (only?).
173  */
174 
175 /*ARGSUSED*/
176 void
177 cachefs_rlent_verify(cachefscache_t *cachep,
178     enum cachefs_rl_type type, uint_t entno)
179 {
180 #ifdef CFSDEBUG
181 	rl_entry_t *rl_ent;
182 	int error;
183 
184 	ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
185 
186 	mutex_enter(&cachep->c_contentslock);
187 
188 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
189 	if (!error && rl_ent->rl_current != type) {
190 #ifdef CFSRLDEBUG
191 		printf("cachefs_rldebug: type should be %x\n", type);
192 		cachefs_rl_debug_show(rl_ent);
193 		debug_enter("cachefs_rlent_verify");
194 #else /* CFSRLDEBUG */
195 		cmn_err(CE_WARN, "rl entry %x type = %x should be %x\n",
196 		    entno, rl_ent->rl_current, type);
197 #endif /* CFSRLDEBUG */
198 	}
199 
200 	mutex_exit(&cachep->c_contentslock);
201 #endif /* CFSDEBUG */
202 }
203 
204 /*
205  * Returns the rl data of the front of the specified resource list.
206  * Returns 0 for success, !0 if the list is empty.
207  */
208 int
209 cachefs_rlent_data(cachefscache_t *cachep, rl_entry_t *valp, uint_t *entnop)
210 {
211 	uint_t entno;
212 	rl_entry_t *rl_ent;
213 	int error = 0;
214 	cachefs_rl_listhead_t *lhp;
215 	enum cachefs_rl_type type;
216 
217 	ASSERT((cachep->c_flags & CACHE_NOCACHE) == 0);
218 
219 	if (entnop == NULL)
220 		entnop = &entno;
221 	*entnop = 0;
222 
223 	mutex_enter(&cachep->c_contentslock);
224 
225 	type = valp->rl_current;
226 	ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
227 	lhp = RL_HEAD(cachep, type);
228 	entno = lhp->rli_front;
229 
230 	if (*entnop == 0) {
231 		error = ENOENT;
232 	} else {
233 		error = cachefs_rl_entry_get(cachep, *entnop, &rl_ent);
234 		if (!error)
235 			*valp = *rl_ent;
236 	}
237 	mutex_exit(&cachep->c_contentslock);
238 	return (error);
239 }
240 
241 /*
242  * This function plucks a slot from the RL free list and creates an RL entry.
243  */
244 int
245 cachefs_rl_alloc(struct cachefscache *cachep, rl_entry_t *valp, uint_t *entnop)
246 {
247 	int error = 0;
248 	uint_t entno;
249 	rl_entry_t *rl_ent;
250 	cachefs_rl_listhead_t *lhp;
251 
252 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
253 	mutex_enter(&cachep->c_contentslock);
254 
255 	cachefs_cache_dirty(cachep, 0);
256 	lhp = RL_HEAD(cachep, CACHEFS_RL_FREE);
257 	entno = lhp->rli_front;
258 	if (entno == 0) {
259 		if (cachep->c_rlinfo.rl_entries >=
260 			cachep->c_label.cl_maxinodes) {
261 			error = ENOMEM;
262 			goto out;
263 		}
264 		entno = ++(cachep->c_rlinfo.rl_entries);
265 		lhp->rli_itemcnt++;
266 		error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
267 		if (error)
268 			goto out;
269 		rl_ent->rl_current = CACHEFS_RL_NONE;
270 		rl_ent->rl_fwd_idx = 0;
271 		rl_ent->rl_bkwd_idx = 0;
272 	}
273 
274 	cachefs_rlent_moveto_nolock(cachep, CACHEFS_RL_NONE, entno, 0);
275 
276 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
277 	if (error)
278 		goto out;
279 	rl_ent->rl_fsid = valp->rl_fsid;
280 	rl_ent->rl_fileno = valp->rl_fileno;
281 	rl_ent->rl_local = valp->rl_local;
282 	rl_ent->rl_attrc = valp->rl_attrc;
283 	rl_ent->rl_fsck = 0;
284 out:
285 	mutex_exit(&cachep->c_contentslock);
286 	if (error == 0)
287 		*entnop = entno;
288 	return (error);
289 }
290 
291 /*
292  * Call to change a local fileno in an rl entry to a normal fileno.
293  */
294 void
295 cachefs_rl_changefileno(cachefscache_t *cachep, uint_t entno, ino64_t fileno)
296 {
297 	rl_entry_t *rl_ent;
298 	int error;
299 
300 	mutex_enter(&cachep->c_contentslock);
301 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
302 	if (!error) {
303 		ASSERT(rl_ent->rl_local);
304 		rl_ent->rl_local = 0;
305 		rl_ent->rl_fileno = fileno;
306 	}
307 	mutex_exit(&cachep->c_contentslock);
308 }
309 
310 /*
311  * Moves the files on the modified list for this file system to
312  * the modified fix list.
313  */
314 void
315 cachefs_move_modified_to_mf(cachefscache_t *cachep, fscache_t *fscp)
316 {
317 	rl_entry_t *list_ent;
318 	uint_t curp, nextp;
319 	cachefs_rl_listhead_t *lhp;
320 	int error;
321 
322 	ASSERT(MUTEX_HELD(&cachep->c_mflock));
323 
324 	mutex_enter(&cachep->c_contentslock);
325 
326 	lhp = RL_HEAD(cachep, CACHEFS_RL_MF);
327 	ASSERT(lhp->rli_front == 0);
328 	ASSERT(lhp->rli_back == 0);
329 	ASSERT(lhp->rli_itemcnt == 0);
330 	lhp->rli_blkcnt = 0;
331 
332 	cachefs_cache_dirty(cachep, 0);
333 
334 	/* walk the modified list */
335 	lhp = RL_HEAD(cachep, CACHEFS_RL_MODIFIED);
336 	for (curp = lhp->rli_front; curp != 0; curp = nextp) {
337 		/* get the next element */
338 		error = cachefs_rl_entry_get(cachep, curp, &list_ent);
339 		if (error) {
340 			mutex_exit(&cachep->c_contentslock);
341 			return;
342 		}
343 		nextp = list_ent->rl_bkwd_idx;
344 
345 		/* skip if element is not in this file system */
346 		if (list_ent->rl_fsid != fscp->fs_cfsid)
347 			continue;
348 
349 		/* move from modified list to mf list */
350 		cachefs_rlent_moveto_nolock(cachep, CACHEFS_RL_MF, curp, 0);
351 	}
352 	mutex_exit(&cachep->c_contentslock);
353 }
354 
355 /*
356  * Moves the contents of the active list to the rl list.
357  * Leave modified files on the active list, so they are not
358  * garbage collected.
359  */
360 void
361 cachefs_rl_cleanup(cachefscache_t *cachep)
362 {
363 	cachefs_rl_listhead_t *lhp;
364 	rl_entry_t *rlp;
365 	uint_t entno, next;
366 	int error;
367 
368 	ASSERT(MUTEX_HELD(&cachep->c_contentslock));
369 
370 	/*
371 	 * if fsck ran, then both of these lists should be empty.  the
372 	 * only time this isn't the case is when we've done a cachefs
373 	 * boot with a clean cache.  then, the cache may have been
374 	 * clean, but files and attrfiles were left dangling.
375 	 *
376 	 * when this happens, we just fix the linked lists here.  this
377 	 * means that the attrcache header and cnode metadata might
378 	 * have incorrect information about which resource lists an
379 	 * entity is currently on.  so, we set CACHE_CHECK_RLTYPE,
380 	 * which says cache-wide to double-check and go with whatever
381 	 * is in the resource list at the time such an object is
382 	 * loaded into memory.
383 	 */
384 
385 	lhp = RL_HEAD(cachep, CACHEFS_RL_ACTIVE);
386 	if (lhp->rli_itemcnt > 0) {
387 		cachep->c_flags |= CACHE_CHECK_RLTYPE;
388 		cachefs_cache_dirty(cachep, 0);
389 	}
390 	for (entno = lhp->rli_front; entno != 0; entno = next) {
391 		error = cachefs_rl_entry_get(cachep, entno, &rlp);
392 		if (error)
393 			return;
394 		next = rlp->rl_bkwd_idx;
395 
396 		ASSERT(rlp->rl_current == CACHEFS_RL_ACTIVE);
397 		cachefs_rlent_moveto_nolock(cachep, CACHEFS_RL_GC, entno, 0);
398 	}
399 
400 #if 0
401 	lhp = RL_HEAD(cachep, CACHEFS_RL_ATTRFILE);
402 	if (lhp->rli_itemcnt > 0) {
403 		cachep->c_flags |= CACHE_CHECK_RLTYPE;
404 		cachefs_cache_dirty(cachep, 0);
405 	}
406 	for (entno = lhp->rli_front; entno != 0; entno = next) {
407 		error = cachefs_rl_entry_get(cachep, entno, &rlp);
408 		if (error)
409 			return;
410 		next = rlp->rl_bkwd_idx;
411 
412 		ASSERT(rlp->rl_current == CACHEFS_RL_ATTRFILE);
413 		cachefs_rlent_moveto_nolock(cachep, CACHEFS_RL_GC, entno, 0);
414 	}
415 #endif
416 }
417 
418 int
419 cachefs_allocfile(cachefscache_t *cachep)
420 {
421 	int error = 0;
422 	int collect = 0;
423 	struct statvfs64 sb;
424 	fsfilcnt64_t used;
425 
426 	(void) VFS_STATVFS(cachep->c_dirvp->v_vfsp, &sb);
427 	used = sb.f_files - sb.f_ffree;
428 
429 	mutex_enter(&cachep->c_contentslock);
430 
431 	/* if there are no more available inodes */
432 	if ((cachep->c_usage.cu_filesused >= cachep->c_label.cl_maxinodes) ||
433 	    ((cachep->c_usage.cu_filesused > cachep->c_label.cl_filemin) &&
434 	    (used > cachep->c_label.cl_filetresh))) {
435 		error = ENOSPC;
436 		if ((cachep->c_flags & CACHE_GARBAGE_COLLECT) == 0)
437 			collect = 1;
438 	}
439 
440 	/* else if there are more available inodes */
441 	else {
442 		cachefs_cache_dirty(cachep, 0);
443 		cachep->c_usage.cu_filesused++;
444 		if (((cachep->c_flags & CACHE_GARBAGE_COLLECT) == 0) &&
445 		    (cachep->c_usage.cu_filesused >=
446 		    cachep->c_label.cl_filehiwat))
447 			collect = 1;
448 	}
449 
450 	mutex_exit(&cachep->c_contentslock);
451 
452 	if (collect)
453 		cachefs_garbage_collect_queue(cachep);
454 
455 	return (error);
456 }
457 
458 void
459 cachefs_freefile(cachefscache_t *cachep)
460 {
461 	mutex_enter(&cachep->c_contentslock);
462 	ASSERT(cachep->c_usage.cu_filesused > 0);
463 	cachefs_cache_dirty(cachep, 0);
464 	cachep->c_usage.cu_filesused--;
465 	mutex_exit(&cachep->c_contentslock);
466 }
467 
468 /*ARGSUSED*/
469 int
470 cachefs_allocblocks(cachefscache_t *cachep, size_t nblks,
471     enum cachefs_rl_type type)
472 {
473 	int error = 0;
474 	int collect = 0;
475 	struct statvfs64 sb;
476 	size_t used;
477 	size_t blocks;
478 
479 	ASSERT(type != CACHEFS_RL_FREE);
480 
481 	(void) VFS_STATVFS(cachep->c_dirvp->v_vfsp, &sb);
482 	used = ((sb.f_blocks - sb.f_bfree) * sb.f_frsize) / MAXBSIZE;
483 
484 	mutex_enter(&cachep->c_contentslock);
485 
486 	/* if there are no more available blocks */
487 	blocks = cachep->c_usage.cu_blksused + nblks;
488 	if ((blocks >= cachep->c_label.cl_maxblks) ||
489 	    ((blocks > cachep->c_label.cl_blockmin) &&
490 	    (used > cachep->c_label.cl_blocktresh))) {
491 		error = ENOSPC;
492 		if ((cachep->c_flags & CACHE_GARBAGE_COLLECT) == 0)
493 			collect = 1;
494 	}
495 
496 	/* else if there are more available blocks */
497 	else {
498 		cachefs_cache_dirty(cachep, 0);
499 		cachep->c_usage.cu_blksused += (uint_t)nblks;
500 		ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
501 		RL_HEAD(cachep, type)->rli_blkcnt += nblks;
502 
503 		if (((cachep->c_flags & CACHE_GARBAGE_COLLECT) == 0) &&
504 		    (cachep->c_usage.cu_blksused >=
505 		    cachep->c_label.cl_blkhiwat))
506 			collect = 1;
507 	}
508 
509 	mutex_exit(&cachep->c_contentslock);
510 
511 	if (collect)
512 		cachefs_garbage_collect_queue(cachep);
513 
514 	return (error);
515 }
516 
517 void
518 cachefs_freeblocks(cachefscache_t *cachep, size_t nblks,
519 		enum cachefs_rl_type type)
520 {
521 	mutex_enter(&cachep->c_contentslock);
522 	cachefs_cache_dirty(cachep, 0);
523 	cachep->c_usage.cu_blksused -= (uint_t)nblks;
524 	ASSERT(cachep->c_usage.cu_blksused >= 0);
525 	ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
526 	ASSERT(type != CACHEFS_RL_FREE);
527 	RL_HEAD(cachep, type)->rli_blkcnt -= nblks;
528 	mutex_exit(&cachep->c_contentslock);
529 }
530 
531 int
532 cachefs_victim(cachefscache_t *cachep)
533 {
534 	uint_t entno;
535 	rl_entry_t *rl_ent;
536 	int error = 0;
537 	ino64_t fsid;
538 	cfs_cid_t cid;
539 	struct fscache *fscp;
540 	struct filegrp *fgp;
541 	struct cachefs_metadata md;
542 	struct cnode *cp;
543 	int isattrc;
544 	cachefs_rl_listhead_t *lhp;
545 
546 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
547 	fscp = NULL;
548 	fgp = NULL;
549 
550 	/* get the file and fsid of the first item on the rl list */
551 	/* XXX call rlent_data() instead */
552 	mutex_enter(&cachep->c_contentslock);
553 	lhp = RL_HEAD(cachep, CACHEFS_RL_GC);
554 	entno = lhp->rli_front;
555 	if (entno == 0) {
556 		mutex_exit(&cachep->c_contentslock);
557 		error = ENOSPC;
558 		goto out;
559 	}
560 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
561 	if (error) {
562 		mutex_exit(&cachep->c_contentslock);
563 		goto out;
564 	}
565 	fsid = rl_ent->rl_fsid;
566 	cid.cid_fileno = rl_ent->rl_fileno;
567 	ASSERT(rl_ent->rl_local == 0);
568 	cid.cid_flags = 0;
569 	isattrc = rl_ent->rl_attrc;
570 	mutex_exit(&cachep->c_contentslock);
571 
572 	/* get the file system cache object for this fsid */
573 	mutex_enter(&cachep->c_fslistlock);
574 	fscp = fscache_list_find(cachep, fsid);
575 	if (fscp == NULL) {
576 		fscp = fscache_create(cachep);
577 		error = fscache_activate(fscp, fsid, NULL, NULL, 0);
578 		if (error) {
579 			cmn_err(CE_WARN,
580 			    "cachefs: cache corruption, run fsck\n");
581 			fscache_destroy(fscp);
582 			fscp = NULL;
583 			mutex_exit(&cachep->c_fslistlock);
584 			error = 0;
585 			goto out;
586 		}
587 		fscache_list_add(cachep, fscp);
588 	}
589 	fscache_hold(fscp);
590 	mutex_exit(&cachep->c_fslistlock);
591 
592 	/* get the file group object for this file */
593 	mutex_enter(&fscp->fs_fslock);
594 	fgp = filegrp_list_find(fscp, &cid);
595 	if (fgp == NULL) {
596 		fgp = filegrp_create(fscp, &cid);
597 		filegrp_list_add(fscp, fgp);
598 	}
599 	if (fgp->fg_flags & CFS_FG_ALLOC_ATTR) {
600 		if (isattrc == 0) {
601 			cmn_err(CE_WARN,
602 			    "cachefs: cache corruption, run fsck\n");
603 			delay(5*hz);
604 		}
605 		filegrp_list_remove(fscp, fgp);
606 		filegrp_destroy(fgp);
607 		error = 0;
608 		fgp = NULL;
609 		mutex_exit(&fscp->fs_fslock);
610 		goto out;
611 	}
612 
613 	/* if we are victimizing an attrcache file */
614 	if (isattrc) {
615 		mutex_enter(&fgp->fg_mutex);
616 		/* if the filegrp is not writable */
617 		if ((fgp->fg_flags & CFS_FG_WRITE) == 0) {
618 			mutex_exit(&fgp->fg_mutex);
619 			error = EROFS;
620 			fgp = NULL;
621 			mutex_exit(&fscp->fs_fslock);
622 			goto out;
623 		}
624 
625 		/* if the filegrp did not go active on us */
626 		if ((fgp->fg_count == 0) && (fgp->fg_header->ach_nffs == 0)) {
627 			mutex_exit(&fgp->fg_mutex);
628 			filegrp_list_remove(fscp, fgp);
629 			fgp->fg_header->ach_count = 0;
630 			filegrp_destroy(fgp);
631 		} else {
632 #ifdef CFSDEBUG
633 			CFS_DEBUG(CFSDEBUG_RESOURCE)
634 				printf("c_victim: filegrp went active"
635 				    " %p %llu %d %d %lld\n",
636 				    (void *) fgp,
637 				    (u_longlong_t)fgp->fg_id.cid_fileno,
638 				    fgp->fg_header->ach_rlno,
639 				    fgp->fg_count, fgp->fg_header->ach_nffs);
640 #endif
641 			ASSERT(fgp->fg_header->ach_rl_current !=
642 			    CACHEFS_RL_GC);
643 			mutex_exit(&fgp->fg_mutex);
644 		}
645 		fgp = NULL;
646 		error = 0;
647 		mutex_exit(&fscp->fs_fslock);
648 		goto out;
649 	}
650 	ASSERT((fgp->fg_flags & CFS_FG_ALLOC_FILE) == 0);
651 	filegrp_hold(fgp);
652 	mutex_exit(&fscp->fs_fslock);
653 
654 	/* grab the cnode list lock */
655 	mutex_enter(&fgp->fg_cnodelock);
656 
657 	/* see if a cnode exists for this file */
658 	(void) cachefs_cnode_find(fgp, &cid, NULL, &cp, NULL, NULL);
659 	if (cp) {
660 		VN_HOLD(CTOV(cp));
661 
662 		/* move file from rl to active list */
663 		cachefs_rlent_moveto(fscp->fs_cache,
664 		    CACHEFS_RL_ACTIVE, cp->c_metadata.md_rlno,
665 		    cp->c_metadata.md_frontblks);
666 		cp->c_metadata.md_rltype = CACHEFS_RL_ACTIVE;
667 		mutex_exit(&cp->c_statelock);
668 		mutex_exit(&fgp->fg_cnodelock);
669 		VN_RELE(CTOV(cp));
670 		error = 0;
671 		goto out;
672 	}
673 
674 	/*
675 	 * The cnode does not exist and since we hold the hashlock
676 	 * it cannot be created until we are done.
677 	 */
678 
679 	/* see if the item is no longer on the rl list, it could happen */
680 	mutex_enter(&cachep->c_contentslock);
681 	lhp = RL_HEAD(cachep, CACHEFS_RL_GC);
682 	entno = lhp->rli_front;
683 	if (entno == 0) {
684 		mutex_exit(&cachep->c_contentslock);
685 		mutex_exit(&fgp->fg_cnodelock);
686 		error = ENOSPC;
687 		goto out;
688 	}
689 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
690 	if (error) {
691 		mutex_exit(&cachep->c_contentslock);
692 		mutex_exit(&fgp->fg_cnodelock);
693 		goto out;
694 	}
695 	if ((fsid != rl_ent->rl_fsid) ||
696 	    (cid.cid_fileno != rl_ent->rl_fileno)) {
697 		mutex_exit(&cachep->c_contentslock);
698 		mutex_exit(&fgp->fg_cnodelock);
699 		error = 0;
700 		goto out;
701 	}
702 	mutex_exit(&cachep->c_contentslock);
703 
704 	/* Get the metadata from the attrcache file */
705 	ASSERT((fgp->fg_flags & CFS_FG_ALLOC_ATTR) == 0);
706 	error = filegrp_read_metadata(fgp, &cid, &md);
707 	ASSERT(error == 0);
708 
709 	/* md.md_rltype may be incorrect, but we know file isn't active. */
710 	if (error) {
711 		/* XXX this should never happen, fix on panic */
712 		mutex_exit(&fgp->fg_cnodelock);
713 		error = 0;
714 		goto out;
715 	}
716 
717 	/* destroy the frontfile */
718 	cachefs_removefrontfile(&md, &cid, fgp);
719 
720 	/* remove the victim from the gc list */
721 	cachefs_rlent_moveto(fscp->fs_cache, CACHEFS_RL_FREE, entno, 0);
722 
723 	/* destroy the metadata */
724 	(void) filegrp_destroy_metadata(fgp, &cid);
725 
726 	mutex_exit(&fgp->fg_cnodelock);
727 	error = 0;
728 out:
729 	if (fgp) {
730 		filegrp_rele(fgp);
731 	}
732 	if (fscp) {
733 		fscache_rele(fscp);
734 	}
735 	return (error);
736 }
737 
738 static void
739 cachefs_garbage_collect(cachefscache_t *cachep)
740 {
741 	fsfilcnt64_t filelowat, filelowatmax, maxfiles, threshfiles;
742 	fsblkcnt64_t blocklowat, blocklowatmax, maxblks, threshblks;
743 	int error;
744 	struct cache_usage *cup = &cachep->c_usage;
745 
746 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
747 	mutex_enter(&cachep->c_contentslock);
748 	ASSERT(cachep->c_flags & CACHE_GARBAGE_COLLECT);
749 	filelowat = cachep->c_label.cl_filelowat;
750 	blocklowat = cachep->c_label.cl_blklowat;
751 	maxblks = cachep->c_label.cl_maxblks;
752 	maxfiles = cachep->c_label.cl_maxinodes;
753 	threshblks = cachep->c_label.cl_blocktresh;
754 	threshfiles = cachep->c_label.cl_filetresh;
755 	mutex_exit(&cachep->c_contentslock);
756 
757 	cachep->c_gc_count++;
758 	cachep->c_gc_time = time;
759 	cachep->c_gc_before = cachefs_gc_front_atime(cachep);
760 
761 	/*
762 	 * since we're here, we're running out of blocks or files.
763 	 * file and block lowat are what determine how low we garbage
764 	 * collect.  in order to do any good, we should drop below
765 	 * maxblocks, threshblocks, or the current blocks, whichever
766 	 * is smaller (same goes for files).  however, we won't go
767 	 * below an arbitrary (small) minimum for each.
768 	 */
769 
770 	/* move down for maxfiles and maxblocks */
771 	if ((filelowatmax = (maxfiles * 7) / 10) < filelowat)
772 		filelowat = filelowatmax;
773 	if ((blocklowatmax = (maxblks * 7) / 10) < blocklowat)
774 		blocklowat = blocklowatmax;
775 
776 	/* move down for threshfiles and threshblocks */
777 	if ((filelowatmax = (threshfiles * 7) / 10) < filelowat)
778 		filelowat = filelowatmax;
779 	if ((blocklowatmax = (threshblks * 7) / 10) < blocklowat)
780 		blocklowat = blocklowatmax;
781 
782 	/* move down for current files and blocks */
783 	if ((filelowatmax = ((fsfilcnt64_t)cup->cu_filesused * 7) / 10) <
784 	    filelowat)
785 		filelowat = filelowatmax;
786 	if ((blocklowatmax = ((fsblkcnt64_t)cup->cu_blksused * 7) / 10) <
787 	    blocklowat)
788 		blocklowat = blocklowatmax;
789 
790 	/* move up for an arbitrary minimum */
791 #define	MIN_BLKLO	640		/* 640*8192 == 5MB */
792 #define	MIN_FILELO	1000
793 	if (filelowat < MIN_FILELO)
794 		filelowat = MIN_FILELO;
795 	if (blocklowat < MIN_BLKLO)
796 		blocklowat = MIN_BLKLO;
797 
798 	while (cup->cu_filesused > filelowat || cup->cu_blksused > blocklowat) {
799 		/* if the thread is to terminate */
800 		if (cachep->c_flags & CACHE_CACHEW_THREADEXIT)
801 			break;
802 
803 		error = cachefs_victim(cachep);
804 		if (error)
805 			break;
806 	}
807 
808 	cachep->c_gc_after = cachefs_gc_front_atime(cachep);
809 	CACHEFS_TIME_TO_CFS_TIME_COPY(cachep->c_gc_after,
810 			cachep->c_rlinfo.rl_gctime, error);
811 }
812 
813 /*
814  * traverse the packed pending list, repacking files when possible.
815  */
816 
817 static void
818 cachefs_packed_pending(cachefscache_t *cachep)
819 {
820 	rl_entry_t rl;
821 	int error = 0; /* not returned -- used as placeholder */
822 	fscache_t *fscp = NULL;
823 	cfs_cid_t cid;
824 	cnode_t *cp;
825 	uint_t entno;
826 	int count = 0;
827 	cachefs_rl_listhead_t *lhp;
828 
829 	ASSERT(MUTEX_HELD(&cachep->c_contentslock));
830 
831 	lhp = RL_HEAD(cachep, CACHEFS_RL_PACKED_PENDING);
832 	count = lhp->rli_itemcnt;
833 
834 	mutex_exit(&cachep->c_contentslock);
835 
836 	rl.rl_current = CACHEFS_RL_PACKED_PENDING;
837 	while (cachefs_rlent_data(cachep, &rl, &entno) == 0) {
838 		if (count-- <= 0) {
839 #ifdef CFSDEBUG
840 			CFS_DEBUG(CFSDEBUG_RESOURCE)
841 				printf("cachefs_ppending: count exceeded\n");
842 #endif /* CFSDEBUG */
843 			break;
844 		}
845 		if ((cachep->c_flags &
846 		    (CACHE_PACKED_PENDING | CACHE_CACHEW_THREADEXIT)) !=
847 		    CACHE_PACKED_PENDING) {
848 #ifdef CFSDEBUG
849 			CFS_DEBUG(CFSDEBUG_RESOURCE)
850 				printf("cachefs_ppending: early exit\n");
851 #endif /* CFSDEBUG */
852 			break;
853 		}
854 		if (rl.rl_current != CACHEFS_RL_PACKED_PENDING) {
855 #ifdef CFSDEBUG
856 			CFS_DEBUG(CFSDEBUG_RESOURCE)
857 				printf("cachefs_ppending: gone from list\n");
858 #endif /* CFSDEBUG */
859 			break;
860 		}
861 
862 		/* if the fscp we have does not match */
863 		if ((fscp == NULL) || (fscp->fs_cfsid != rl.rl_fsid)) {
864 			if (fscp) {
865 				cachefs_cd_release(fscp);
866 				fscache_rele(fscp);
867 				fscp = NULL;
868 			}
869 
870 			/* get the file system cache object for this fsid */
871 			mutex_enter(&cachep->c_fslistlock);
872 			fscp = fscache_list_find(cachep, rl.rl_fsid);
873 			if (fscp == NULL) {
874 
875 				/*
876 				 * uh oh, the filesystem probably
877 				 * isn't mounted.  we `move' this
878 				 * entry onto the same list that it's
879 				 * on, which really just moves it to
880 				 * the back of the list.  we need not
881 				 * worry about an infinite loop, due
882 				 * to the counter.
883 				 */
884 
885 				cachefs_rlent_moveto(cachep,
886 				    CACHEFS_RL_PACKED_PENDING, entno, 0);
887 #ifdef CFSDEBUG
888 				CFS_DEBUG(CFSDEBUG_RESOURCE)
889 					printf("cachefs_ppending: "
890 					    "fscp find failed\n");
891 #endif /* CFSDEBUG */
892 				continue;
893 			}
894 			fscache_hold(fscp);
895 			mutex_exit(&cachep->c_fslistlock);
896 
897 			/* get access to the file system */
898 			error = cachefs_cd_access(fscp, 0, 0);
899 			if ((error) ||
900 			    (fscp->fs_cdconnected != CFS_CD_CONNECTED)) {
901 #ifdef CFSDEBUG
902 				CFS_DEBUG(CFSDEBUG_RESOURCE)
903 					printf("cachefs: "
904 					    "ppending: err %d con %d\n",
905 					    error, fscp->fs_cdconnected);
906 #endif /* CFSDEBUG */
907 				fscache_rele(fscp);
908 				fscp = NULL;
909 				break;
910 			}
911 		}
912 
913 		/* get the cnode for the file */
914 		cid.cid_fileno = rl.rl_fileno;
915 		cid.cid_flags = rl.rl_local ? CFS_CID_LOCAL : 0;
916 		error = cachefs_cnode_make(&cid, fscp,
917 		    NULL, NULL, NULL, kcred, 0, &cp);
918 		if (error) {
919 #ifdef CFSDEBUG
920 			CFS_DEBUG(CFSDEBUG_RESOURCE)
921 				printf("cachefs: "
922 				    "ppending: could not find %llu\n",
923 				    (u_longlong_t)cid.cid_fileno);
924 			delay(5*hz);
925 #endif /* CFSDEBUG */
926 			break;
927 		}
928 
929 		mutex_enter(&cp->c_statelock);
930 		if (cp->c_flags & CN_STALE) {
931 			/* back file went away behind our back */
932 			ASSERT(cp->c_metadata.md_rlno == 0);
933 			mutex_exit(&cp->c_statelock);
934 
935 #ifdef CFSDEBUG
936 			CFS_DEBUG(CFSDEBUG_RESOURCE)
937 				printf("cachefs: ppending: stale\n");
938 #endif /* CFSDEBUG */
939 
940 			VN_RELE(CTOV(cp));
941 			continue;
942 		}
943 		mutex_exit(&cp->c_statelock);
944 
945 		error = cachefs_pack_common(CTOV(cp),
946 		    (cp->c_cred) ? cp->c_cred : kcred);
947 		VN_RELE(CTOV(cp));
948 
949 		if (error != 0) {
950 #ifdef CFSDEBUG
951 			CFS_DEBUG(CFSDEBUG_RESOURCE)
952 				printf("cachefs: "
953 				    "ppending: pack_common: error = %d\n",
954 				    error);
955 #endif /* CFSDEBUG */
956 			break;
957 		}
958 	}
959 
960 	if (fscp != NULL) {
961 		cachefs_cd_release(fscp);
962 		fscache_rele(fscp);
963 	}
964 
965 	mutex_enter(&cachep->c_contentslock);
966 	if (lhp->rli_itemcnt == 0)
967 		cachep->c_flags &= ~CACHE_PACKED_PENDING;
968 }
969 
970 /* seconds; interval to do ppend list */
971 static time_t cachefs_ppend_time = 900;
972 
973 /* main routine for the cachep worker thread */
974 void
975 cachefs_cachep_worker_thread(cachefscache_t *cachep)
976 {
977 	int error;
978 	struct flock64 fl;
979 	callb_cpr_t cprinfo;
980 	kmutex_t cpr_lock;
981 
982 
983 	/* lock the lock file for exclusive write access */
984 	fl.l_type = F_WRLCK;
985 	fl.l_whence = 0;
986 	fl.l_start = (offset_t)0;
987 	fl.l_len = (offset_t)1024;
988 	fl.l_sysid = 0;
989 	fl.l_pid = 0;
990 	error = VOP_FRLOCK(cachep->c_lockvp, F_SETLK, &fl, FWRITE,
991 		(offset_t)0, NULL, kcred, NULL);
992 	if (error) {
993 		cmn_err(CE_WARN,
994 		    "cachefs: Can't lock Cache Lock File(r); Error %d\n",
995 		    error);
996 	}
997 
998 	mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL);
999 	CALLB_CPR_INIT(&cprinfo, &cpr_lock, callb_generic_cpr, "cfs_gct");
1000 	mutex_enter(&cpr_lock);
1001 	mutex_enter(&cachep->c_contentslock);
1002 
1003 	/* loop while the thread is allowed to run */
1004 	while ((cachep->c_flags & CACHE_CACHEW_THREADEXIT) == 0) {
1005 		clock_t wakeup;
1006 
1007 		/* wait for a wakeup call */
1008 		cachep->c_flags &= ~CACHE_GARBAGE_COLLECT;
1009 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
1010 		mutex_exit(&cpr_lock);
1011 		wakeup = (clock_t)(lbolt + (cachefs_ppend_time * hz));
1012 		(void) cv_timedwait(&cachep->c_cwcv,
1013 			&cachep->c_contentslock, wakeup);
1014 		mutex_enter(&cpr_lock);
1015 		CALLB_CPR_SAFE_END(&cprinfo, &cpr_lock);
1016 
1017 		/* if the thread is to terminate */
1018 		if (cachep->c_flags & CACHE_CACHEW_THREADEXIT)
1019 			break;
1020 
1021 		/* thread is running during nofill, but just to hold lock */
1022 		if (cachep->c_flags & CACHE_NOFILL)
1023 			continue;
1024 
1025 		/* if garbage collection is to run */
1026 		if (cachep->c_flags & CACHE_GARBAGE_COLLECT) {
1027 			mutex_exit(&cachep->c_contentslock);
1028 			cachefs_garbage_collect(cachep);
1029 
1030 			/*
1031 			 * Prevent garbage collection from running more
1032 			 * than once every 30 seconds.  This addresses
1033 			 * those cases which do not allow removing
1034 			 * an item from the rl by keeping gc from
1035 			 * being a spin loop.
1036 			 */
1037 			delay(30*hz); /* XXX sam: still do this? */
1038 			mutex_enter(&cachep->c_contentslock);
1039 		}
1040 
1041 		if (cachep->c_flags & CACHE_PACKED_PENDING)
1042 			cachefs_packed_pending(cachep);
1043 		ASSERT(MUTEX_HELD(&cachep->c_contentslock));
1044 	}
1045 
1046 	cachep->c_flags &= ~CACHE_CACHEW_THREADRUN;
1047 	cv_broadcast(&cachep->c_cwhaltcv);
1048 	CALLB_CPR_EXIT(&cprinfo);
1049 	mutex_exit(&cachep->c_contentslock);
1050 	mutex_destroy(&cpr_lock);
1051 
1052 	/* unlock the lock file */
1053 	fl.l_type = F_UNLCK;
1054 	fl.l_whence = 0;
1055 	fl.l_start = (offset_t)0;
1056 	fl.l_len = (offset_t)1024;
1057 	fl.l_sysid = 0;
1058 	fl.l_pid = 0;
1059 	error = VOP_FRLOCK(cachep->c_lockvp, F_SETLK, &fl,
1060 		FWRITE, (offset_t)0, NULL, kcred, NULL);
1061 	if (error) {
1062 		cmn_err(CE_WARN, "cachefs: Can't unlock lock file\n");
1063 	}
1064 
1065 	thread_exit();
1066 	/*NOTREACHED*/
1067 }
1068 
1069 /* queues up a request to run the garbage collection */
1070 void
1071 cachefs_garbage_collect_queue(cachefscache_t *cachep)
1072 {
1073 	cachefs_rl_listhead_t *lhp;
1074 
1075 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
1076 	mutex_enter(&cachep->c_contentslock);
1077 
1078 	/* quit if there is no garbage collection thread */
1079 	if ((cachep->c_flags & CACHE_CACHEW_THREADRUN) == 0) {
1080 		mutex_exit(&cachep->c_contentslock);
1081 		return;
1082 	}
1083 
1084 	/* quit if garbage collection is already in progress */
1085 	if (cachep->c_flags & CACHE_GARBAGE_COLLECT) {
1086 		mutex_exit(&cachep->c_contentslock);
1087 		return;
1088 	}
1089 
1090 	/* quit if there is no garbage to collect */
1091 	lhp = RL_HEAD(cachep, CACHEFS_RL_GC);
1092 	if (lhp->rli_front == 0) {
1093 		mutex_exit(&cachep->c_contentslock);
1094 		return;
1095 	}
1096 
1097 	/* indicate garbage collecting is in progress */
1098 	cachep->c_flags |= CACHE_GARBAGE_COLLECT;
1099 
1100 	/* wake up the garbage collection thread */
1101 	cv_signal(&cachep->c_cwcv);
1102 
1103 	mutex_exit(&cachep->c_contentslock);
1104 }
1105 
1106 #ifdef CFSRLDEBUG
1107 time_t cachefs_dbvalid = 123; /* default to non-zero junk */
1108 struct kmem_cache *cachefs_rl_debug_cache = NULL;
1109 static int cachefs_rl_debug_maxcount = CACHEFS_RLDB_DEF_MAXCOUNT;
1110 kmutex_t cachefs_rl_debug_mutex;
1111 static int cachefs_rl_debug_inuse = 0;
1112 
1113 void
1114 cachefs_rl_debug_reclaim(void *cdrarg)
1115 {
1116 	extern cachefscache_t *cachefs_cachelist;
1117 	cachefscache_t *cachep;
1118 	int index;
1119 	int error;
1120 
1121 	for (cachep = cachefs_cachelist; cachep != NULL;
1122 		cachep = cachep->c_next) {
1123 		mutex_enter(&cachep->c_contentslock);
1124 
1125 		for (index = 0;
1126 		    index <= cachep->c_rlinfo.rl_entries;
1127 		    index++) {
1128 			rl_entry_t *rlent;
1129 
1130 			error = cachefs_rl_entry_get(cachep, index, &rlent);
1131 			if (error)
1132 				break;
1133 			cachefs_rl_debug_destroy(rlent);
1134 		}
1135 
1136 		mutex_exit(&cachep->c_contentslock);
1137 	}
1138 }
1139 
1140 void
1141 cachefs_rl_debug_save(rl_entry_t *rlent)
1142 {
1143 	rl_debug_t *rldb, *prev, *next;
1144 	int count = 0;
1145 
1146 	mutex_enter(&cachefs_rl_debug_mutex);
1147 	if (cachefs_rl_debug_cache == NULL)
1148 		cachefs_rl_debug_cache =
1149 		    kmem_cache_create("cachefs_rl_debug",
1150 			sizeof (rl_debug_t), 0,
1151 			NULL, NULL, cachefs_rl_debug_reclaim, NULL, NULL, 0);
1152 
1153 	rldb = kmem_cache_alloc(cachefs_rl_debug_cache, KM_SLEEP);
1154 	++cachefs_rl_debug_inuse;
1155 
1156 	rldb->db_hrtime = gethrtime();
1157 
1158 	rldb->db_attrc = rlent->rl_attrc;
1159 	rldb->db_fsck = rlent->rl_fsck;
1160 	rldb->db_fsid = rlent->rl_fsid;
1161 	rldb->db_fileno = rlent->rl_fileno;
1162 	rldb->db_current = rlent->rl_current;
1163 
1164 	rldb->db_stackheight = getpcstack(rldb->db_stack,
1165 	    CACHEFS_RLDB_STACKSIZE);
1166 
1167 	if (rlent->rl_dbvalid == cachefs_dbvalid) {
1168 		rldb->db_next = rlent->rl_debug;
1169 	} else {
1170 		rldb->db_next = NULL;
1171 		rlent->rl_dbvalid = cachefs_dbvalid;
1172 	}
1173 	rlent->rl_debug = rldb;
1174 
1175 	prev = rldb;
1176 	for (rldb = rldb->db_next; rldb != NULL; rldb = next) {
1177 		next = rldb->db_next;
1178 		if (++count >= cachefs_rl_debug_maxcount) {
1179 			if (prev != NULL)
1180 				prev->db_next = NULL;
1181 			kmem_cache_free(cachefs_rl_debug_cache, rldb);
1182 			--cachefs_rl_debug_inuse;
1183 			prev = NULL;
1184 		} else {
1185 			prev = rldb;
1186 		}
1187 	}
1188 	mutex_exit(&cachefs_rl_debug_mutex);
1189 }
1190 
1191 void
1192 cachefs_rl_debug_show(rl_entry_t *rlent)
1193 {
1194 	rl_debug_t *rldb;
1195 	hrtime_t now, elapse;
1196 	timestruc_t tv;
1197 	char *cname = NULL;
1198 	int i;
1199 
1200 	mutex_enter(&cachefs_rl_debug_mutex);
1201 	if (rlent->rl_dbvalid != cachefs_dbvalid) {
1202 		printf("cachefs_rldb: rl entry at %lx -- no info!\n",
1203 		    (uintptr_t)rlent);
1204 		mutex_exit(&cachefs_rl_debug_mutex);
1205 		return;
1206 	}
1207 
1208 	now = gethrtime();
1209 	hrt2ts(now, &tv);
1210 
1211 	printf("===== cachefs_rldb start at %ld =====\n", tv.tv_sec);
1212 	printf("-==== i am thread id %lx   ====-\n", (uintptr_t)curthread);
1213 
1214 	for (rldb = rlent->rl_debug;
1215 	    rldb != NULL;
1216 	    rldb = rldb->db_next) {
1217 		printf("----- cachefs_rldb record start -----\n");
1218 		elapse = now - rldb->db_hrtime;
1219 		hrt2ts(elapse, &tv);
1220 		printf("cachefs_rldb: ago = %lds %ldus\n",
1221 		    tv.tv_sec, tv.tv_nsec / 1000);
1222 
1223 		printf("cachefs_rldb: rl_attrc = %d\n", rldb->db_attrc);
1224 		printf("cachefs_rldb: rl_fsck = %d\n", rldb->db_fsck);
1225 		printf("cachefs_rldb: rl_fsid = %u\n", rldb->db_fsid);
1226 		printf("cachefs_rldb: rl_fileno = %lu\n", rldb->db_fileno);
1227 
1228 		switch (rldb->db_current) {
1229 		case CACHEFS_RL_NONE:
1230 			cname = "CACHEFS_RL_NONE";
1231 			break;
1232 		case CACHEFS_RL_FREE:
1233 			cname = "CACHEFS_RL_FREE";
1234 			break;
1235 		case CACHEFS_RL_GC:
1236 			cname = "CACHEFS_RL_GC";
1237 			break;
1238 		case CACHEFS_RL_ACTIVE:
1239 			cname = "CACHEFS_RL_ACTIVE";
1240 			break;
1241 		case CACHEFS_RL_ATTRFILE:
1242 			cname = "CACHEFS_RL_ATTRFILE";
1243 			break;
1244 		case CACHEFS_RL_MODIFIED:
1245 			cname = "CACHEFS_RL_MODIFIED";
1246 			break;
1247 		case CACHEFS_RL_PACKED:
1248 			cname = "CACHEFS_RL_PACKED";
1249 			break;
1250 		case CACHEFS_RL_PACKED_PENDING:
1251 			cname = "CACHEFS_RL_PACKED_PENDING";
1252 			break;
1253 		case CACHEFS_RL_MF:
1254 			cname = "CACHEFS_MF_GC";
1255 			break;
1256 		}
1257 		if (cname != NULL) {
1258 			printf("cachefs_rldb: state = %s\n", cname);
1259 		} else {
1260 			printf("cachefs_rldb: undefined state %x\n",
1261 			    rldb->db_current);
1262 		}
1263 
1264 		printf("cachefs_rldb: stack trace\n");
1265 		for (i = 0; i < rldb->db_stackheight; i++) {
1266 			char *sym;
1267 			uint_t off;
1268 
1269 			sym = kobj_getsymname(rldb->db_stack[i], &off);
1270 			printf("cachefs_rldb:    %s+%lx\n",
1271 			    sym ? sym : "?", off);
1272 			delay(hz/4);
1273 		}
1274 
1275 		printf("----- cachefs_rldb record end -----\n");
1276 	}
1277 
1278 	mutex_exit(&cachefs_rl_debug_mutex);
1279 }
1280 
1281 void
1282 cachefs_rl_debug_destroy(rl_entry_t *rlent)
1283 {
1284 	rl_debug_t *rldb, *next;
1285 
1286 	mutex_enter(&cachefs_rl_debug_mutex);
1287 	if (rlent->rl_dbvalid != cachefs_dbvalid) {
1288 		rlent->rl_debug = NULL;
1289 		mutex_exit(&cachefs_rl_debug_mutex);
1290 		return;
1291 	}
1292 
1293 	for (rldb = rlent->rl_debug; rldb != NULL; rldb = next) {
1294 		next = rldb->db_next;
1295 		kmem_cache_free(cachefs_rl_debug_cache, rldb);
1296 		--cachefs_rl_debug_inuse;
1297 	}
1298 
1299 	rlent->rl_debug = NULL;
1300 	mutex_exit(&cachefs_rl_debug_mutex);
1301 }
1302 #endif /* CFSRLDEBUG */
1303 
1304 int
1305 cachefs_rl_entry_get(cachefscache_t *cachep, uint_t entno, rl_entry_t **ent)
1306 {
1307 	rl_entry_t *rl_ent;
1308 	uint_t whichwindow, winoffset;
1309 	int error = 0;
1310 
1311 	ASSERT(MUTEX_HELD(&cachep->c_contentslock));
1312 	ASSERT(entno <= cachep->c_label.cl_maxinodes); /* strictly less? */
1313 #if 0
1314 	ASSERT((cachep->c_flags & CACHE_NOFILL) == 0);
1315 #endif
1316 
1317 	whichwindow = entno / CACHEFS_RLPMBS;
1318 	winoffset = entno % CACHEFS_RLPMBS;
1319 
1320 	if ((cachep->c_rl_entries == NULL) ||
1321 	    (cachep->c_rl_window != whichwindow)) {
1322 		if (cachep->c_rl_entries != NULL) {
1323 			error = vn_rdwr(UIO_WRITE, cachep->c_resfilevp,
1324 			    (caddr_t)cachep->c_rl_entries, MAXBSIZE,
1325 			    (offset_t)((cachep->c_rl_window + 1) * MAXBSIZE),
1326 			    UIO_SYSSPACE, 0, RLIM_INFINITY, kcred, NULL);
1327 			if (error)
1328 				return (error);
1329 		}
1330 		else
1331 			cachep->c_rl_entries = (rl_entry_t *)
1332 			    cachefs_kmem_alloc(MAXBSIZE, KM_SLEEP);
1333 
1334 		error = vn_rdwr(UIO_READ, cachep->c_resfilevp,
1335 		    (caddr_t)cachep->c_rl_entries, MAXBSIZE,
1336 		    (offset_t)((whichwindow + 1) * MAXBSIZE),
1337 		    UIO_SYSSPACE, 0, RLIM_INFINITY, kcred, NULL);
1338 		if (error) {
1339 			cachefs_kmem_free(cachep->c_rl_entries, MAXBSIZE);
1340 			cachep->c_rl_entries = NULL;
1341 			return (error);
1342 		}
1343 		cachep->c_rl_window = whichwindow;
1344 	}
1345 	rl_ent = &cachep->c_rl_entries[winoffset];
1346 
1347 	*ent = rl_ent;
1348 #ifdef CFSRLDEBUG
1349 	cachefs_rl_debug_save(rl_ent);
1350 #endif /* CFSRLDEBUG */
1351 
1352 	return (error);
1353 }
1354 
1355 static time_t
1356 cachefs_gc_front_atime(cachefscache_t *cachep)
1357 {
1358 	char namebuf[CFS_FRONTFILE_NAME_SIZE];
1359 
1360 	rl_entry_t rl, *rl_ent;
1361 	uint_t entno, fgsize;
1362 	cfs_cid_t dircid, cid;
1363 	struct fscache *fscp;
1364 	cachefs_rl_listhead_t *lhp;
1365 	int error;
1366 
1367 	struct vnode *dirvp, *filevp;
1368 	struct vattr va;
1369 
1370 	int reledir = 0;
1371 	int gotfile = 0;
1372 	time_t rc = (time_t)0;
1373 
1374 	mutex_enter(&cachep->c_contentslock);
1375 	lhp = RL_HEAD(cachep, CACHEFS_RL_GC);
1376 	entno = lhp->rli_front;
1377 	if (entno == 0) {
1378 		mutex_exit(&cachep->c_contentslock);
1379 		goto out;
1380 	}
1381 
1382 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
1383 	if (error) {
1384 		mutex_exit(&cachep->c_contentslock);
1385 		goto out;
1386 	}
1387 	rl = *rl_ent;
1388 	mutex_exit(&cachep->c_contentslock);
1389 	cid.cid_fileno = rl.rl_fileno;
1390 	ASSERT(rl.rl_local == 0);
1391 	cid.cid_flags = 0;
1392 	dircid.cid_flags = 0;
1393 	mutex_enter(&cachep->c_fslistlock);
1394 	if ((fscp = fscache_list_find(cachep, rl.rl_fsid)) == NULL) {
1395 		mutex_exit(&cachep->c_fslistlock);
1396 		goto out;
1397 	}
1398 
1399 	if (rl.rl_attrc) {
1400 		make_ascii_name(&cid, namebuf);
1401 		dirvp = fscp->fs_fsattrdir;
1402 	} else {
1403 		dirvp = NULL;
1404 		fgsize = fscp->fs_info.fi_fgsize;
1405 		dircid.cid_fileno = ((cid.cid_fileno / fgsize) * fgsize);
1406 		make_ascii_name(&dircid, namebuf);
1407 		if (VOP_LOOKUP(fscp->fs_fscdirvp, namebuf,
1408 		    &dirvp, (struct pathname *)NULL, 0,
1409 		    (vnode_t *)NULL, kcred, NULL, NULL, NULL) == 0) {
1410 			make_ascii_name(&cid, namebuf);
1411 			reledir++;
1412 		} else {
1413 			mutex_exit(&cachep->c_fslistlock);
1414 			goto out;
1415 		}
1416 	}
1417 	if (dirvp && VOP_LOOKUP(dirvp, namebuf, &filevp,
1418 	    (struct pathname *)NULL, 0,
1419 	    (vnode_t *)NULL, kcred, NULL, NULL, NULL) == 0) {
1420 		gotfile = 1;
1421 	}
1422 	if (reledir)
1423 		VN_RELE(dirvp);
1424 	mutex_exit(&cachep->c_fslistlock);
1425 
1426 	if (gotfile) {
1427 		va.va_mask = AT_ATIME;
1428 		if (VOP_GETATTR(filevp, &va, 0, kcred, NULL) == 0)
1429 			rc = va.va_atime.tv_sec;
1430 		VN_RELE(filevp);
1431 	}
1432 
1433 out:
1434 	return (rc);
1435 }
1436