xref: /titanic_41/usr/src/uts/common/fs/cachefs/cachefs_resource.c (revision 549ec3fff108310966327d1dc9004551b63210b7)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/param.h>
29 #include <sys/types.h>
30 #include <sys/systm.h>
31 #include <sys/cred.h>
32 #include <sys/proc.h>
33 #include <sys/user.h>
34 #include <sys/vfs.h>
35 #include <sys/vnode.h>
36 #include <sys/pathname.h>
37 #include <sys/uio.h>
38 #include <sys/tiuser.h>
39 #include <sys/sysmacros.h>
40 #include <sys/kmem.h>
41 #include <sys/kobj.h>
42 #include <sys/mount.h>
43 #include <sys/ioctl.h>
44 #include <sys/statvfs.h>
45 #include <sys/errno.h>
46 #include <sys/debug.h>
47 #include <sys/cmn_err.h>
48 #include <sys/utsname.h>
49 #include <sys/modctl.h>
50 #include <sys/file.h>
51 #include <sys/stat.h>
52 #include <sys/fcntl.h>
53 #include <sys/callb.h>
54 
55 #include <vm/hat.h>
56 #include <vm/as.h>
57 #include <vm/page.h>
58 #include <vm/pvn.h>
59 #include <vm/seg.h>
60 #include <vm/seg_map.h>
61 #include <vm/seg_vn.h>
62 #include <vm/rm.h>
63 #include <sys/fs/cachefs_fs.h>
64 
65 extern time_t time;
66 
67 /* forward references */
68 int cachefs_rl_entry_get(cachefscache_t *, uint_t, rl_entry_t **);
69 void cachefs_garbage_collect_queue(cachefscache_t *cachep);
70 static time_t cachefs_gc_front_atime(cachefscache_t *cachep);
71 static void cachefs_garbage_collect(cachefscache_t *cachep);
72 static void cachefs_packed_pending(cachefscache_t *cachep);
73 
74 
75 #define	RL_HEAD(cachep, type) \
76 	(&(cachep->c_rlinfo.rl_items[CACHEFS_RL_INDEX(type)]))
77 
78 /*
79  * This function moves an RL entry from whereever it currently is to
80  * the back of the requested list.
81  */
82 void
83 cachefs_rlent_moveto(cachefscache_t *cachep,
84     enum cachefs_rl_type type, uint_t entno, size_t blks)
85 {
86 	mutex_enter(&cachep->c_contentslock);
87 	cachefs_cache_dirty(cachep, 0);
88 	cachefs_rlent_moveto_nolock(cachep, type, entno, blks);
89 	mutex_exit(&cachep->c_contentslock);
90 }
91 
92 void
93 cachefs_rlent_moveto_nolock(cachefscache_t *cachep,
94     enum cachefs_rl_type type, uint_t entno, size_t blks)
95 {
96 	rl_entry_t *rl_ent;
97 	uint_t prev, next;
98 	cachefs_rl_listhead_t *lhp;
99 	enum cachefs_rl_type otype;
100 	int error;
101 
102 	ASSERT(entno != 0);
103 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
104 	ASSERT(MUTEX_HELD(&cachep->c_contentslock));
105 	ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
106 
107 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
108 	if (error)
109 		return;
110 	next = rl_ent->rl_fwd_idx;
111 	prev = rl_ent->rl_bkwd_idx;
112 	otype = rl_ent->rl_current;
113 	ASSERT((CACHEFS_RL_START <= otype) && (otype <= CACHEFS_RL_END));
114 	rl_ent->rl_current = CACHEFS_RL_NONE;
115 
116 	if (type == CACHEFS_RL_PACKED_PENDING) {
117 		/* XXX sam: is this the right place to turn this on? */
118 		cachep->c_flags |= CACHE_PACKED_PENDING;
119 	}
120 
121 	/* remove entry from its previous list */
122 
123 	lhp = RL_HEAD(cachep, otype);
124 	if ((lhp->rli_back == 0) || (lhp->rli_front == 0))
125 		ASSERT((lhp->rli_back == 0) && (lhp->rli_front == 0));
126 
127 	if (lhp->rli_back == entno)
128 		lhp->rli_back = next;
129 	if (lhp->rli_front == entno)
130 		lhp->rli_front = prev;
131 	if (prev != 0) {
132 		error = cachefs_rl_entry_get(cachep, prev, &rl_ent);
133 		if (error)
134 			return;
135 		rl_ent->rl_fwd_idx = next;
136 	}
137 	if (next != 0) {
138 		error = cachefs_rl_entry_get(cachep, next, &rl_ent);
139 		if (error)
140 			return;
141 		rl_ent->rl_bkwd_idx = prev;
142 	}
143 	lhp->rli_blkcnt -= blks;
144 	lhp->rli_itemcnt--;
145 
146 	/* add entry to its new list */
147 
148 	lhp = RL_HEAD(cachep, type);
149 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
150 	if (error)
151 		return;
152 	rl_ent->rl_current = type;
153 	rl_ent->rl_bkwd_idx = 0;
154 	rl_ent->rl_fwd_idx = lhp->rli_back;
155 
156 	if (lhp->rli_back != 0) {
157 		ASSERT(lhp->rli_front != 0);
158 		error = cachefs_rl_entry_get(cachep, lhp->rli_back, &rl_ent);
159 		if (error)
160 			return;
161 		rl_ent->rl_bkwd_idx = entno;
162 	} else {
163 		ASSERT(lhp->rli_front == 0);
164 		lhp->rli_front = entno;
165 	}
166 	lhp->rli_back = entno;
167 	lhp->rli_blkcnt += blks;
168 	lhp->rli_itemcnt++;
169 }
170 
171 /*
172  * This function verifies that an rl entry is of the `correct' type.
173  * it's used for debugging (only?).
174  */
175 
176 /*ARGSUSED*/
177 void
178 cachefs_rlent_verify(cachefscache_t *cachep,
179     enum cachefs_rl_type type, uint_t entno)
180 {
181 #ifdef CFSDEBUG
182 	rl_entry_t *rl_ent;
183 	int error;
184 
185 	ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
186 
187 	mutex_enter(&cachep->c_contentslock);
188 
189 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
190 	if (!error && rl_ent->rl_current != type) {
191 #ifdef CFSRLDEBUG
192 		printf("cachefs_rldebug: type should be %x\n", type);
193 		cachefs_rl_debug_show(rl_ent);
194 		debug_enter("cachefs_rlent_verify");
195 #else /* CFSRLDEBUG */
196 		cmn_err(CE_WARN, "rl entry %x type = %x should be %x\n",
197 		    entno, rl_ent->rl_current, type);
198 #endif /* CFSRLDEBUG */
199 	}
200 
201 	mutex_exit(&cachep->c_contentslock);
202 #endif /* CFSDEBUG */
203 }
204 
205 /*
206  * Returns the rl data of the front of the specified resource list.
207  * Returns 0 for success, !0 if the list is empty.
208  */
209 int
210 cachefs_rlent_data(cachefscache_t *cachep, rl_entry_t *valp, uint_t *entnop)
211 {
212 	uint_t entno;
213 	rl_entry_t *rl_ent;
214 	int error = 0;
215 	cachefs_rl_listhead_t *lhp;
216 	enum cachefs_rl_type type;
217 
218 	ASSERT((cachep->c_flags & CACHE_NOCACHE) == 0);
219 
220 	if (entnop == NULL)
221 		entnop = &entno;
222 	*entnop = 0;
223 
224 	mutex_enter(&cachep->c_contentslock);
225 
226 	type = valp->rl_current;
227 	ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
228 	lhp = RL_HEAD(cachep, type);
229 	entno = lhp->rli_front;
230 
231 	if (*entnop == 0) {
232 		error = ENOENT;
233 	} else {
234 		error = cachefs_rl_entry_get(cachep, *entnop, &rl_ent);
235 		if (!error)
236 			*valp = *rl_ent;
237 	}
238 	mutex_exit(&cachep->c_contentslock);
239 	return (error);
240 }
241 
242 /*
243  * This function plucks a slot from the RL free list and creates an RL entry.
244  */
245 int
246 cachefs_rl_alloc(struct cachefscache *cachep, rl_entry_t *valp, uint_t *entnop)
247 {
248 	int error = 0;
249 	uint_t entno;
250 	rl_entry_t *rl_ent;
251 	cachefs_rl_listhead_t *lhp;
252 
253 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
254 	mutex_enter(&cachep->c_contentslock);
255 
256 	cachefs_cache_dirty(cachep, 0);
257 	lhp = RL_HEAD(cachep, CACHEFS_RL_FREE);
258 	entno = lhp->rli_front;
259 	if (entno == 0) {
260 		if (cachep->c_rlinfo.rl_entries >=
261 			cachep->c_label.cl_maxinodes) {
262 			error = ENOMEM;
263 			goto out;
264 		}
265 		entno = ++(cachep->c_rlinfo.rl_entries);
266 		lhp->rli_itemcnt++;
267 		error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
268 		if (error)
269 			goto out;
270 		rl_ent->rl_current = CACHEFS_RL_NONE;
271 		rl_ent->rl_fwd_idx = 0;
272 		rl_ent->rl_bkwd_idx = 0;
273 	}
274 
275 	cachefs_rlent_moveto_nolock(cachep, CACHEFS_RL_NONE, entno, 0);
276 
277 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
278 	if (error)
279 		goto out;
280 	rl_ent->rl_fsid = valp->rl_fsid;
281 	rl_ent->rl_fileno = valp->rl_fileno;
282 	rl_ent->rl_local = valp->rl_local;
283 	rl_ent->rl_attrc = valp->rl_attrc;
284 	rl_ent->rl_fsck = 0;
285 out:
286 	mutex_exit(&cachep->c_contentslock);
287 	if (error == 0)
288 		*entnop = entno;
289 	return (error);
290 }
291 
292 /*
293  * Call to change a local fileno in an rl entry to a normal fileno.
294  */
295 void
296 cachefs_rl_changefileno(cachefscache_t *cachep, uint_t entno, ino64_t fileno)
297 {
298 	rl_entry_t *rl_ent;
299 	int error;
300 
301 	mutex_enter(&cachep->c_contentslock);
302 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
303 	if (!error) {
304 		ASSERT(rl_ent->rl_local);
305 		rl_ent->rl_local = 0;
306 		rl_ent->rl_fileno = fileno;
307 	}
308 	mutex_exit(&cachep->c_contentslock);
309 }
310 
311 /*
312  * Moves the files on the modified list for this file system to
313  * the modified fix list.
314  */
315 void
316 cachefs_move_modified_to_mf(cachefscache_t *cachep, fscache_t *fscp)
317 {
318 	rl_entry_t *list_ent;
319 	uint_t curp, nextp;
320 	cachefs_rl_listhead_t *lhp;
321 	int error;
322 
323 	ASSERT(MUTEX_HELD(&cachep->c_mflock));
324 
325 	mutex_enter(&cachep->c_contentslock);
326 
327 	lhp = RL_HEAD(cachep, CACHEFS_RL_MF);
328 	ASSERT(lhp->rli_front == 0);
329 	ASSERT(lhp->rli_back == 0);
330 	ASSERT(lhp->rli_itemcnt == 0);
331 	lhp->rli_blkcnt = 0;
332 
333 	cachefs_cache_dirty(cachep, 0);
334 
335 	/* walk the modified list */
336 	lhp = RL_HEAD(cachep, CACHEFS_RL_MODIFIED);
337 	for (curp = lhp->rli_front; curp != 0; curp = nextp) {
338 		/* get the next element */
339 		error = cachefs_rl_entry_get(cachep, curp, &list_ent);
340 		if (error) {
341 			mutex_exit(&cachep->c_contentslock);
342 			return;
343 		}
344 		nextp = list_ent->rl_bkwd_idx;
345 
346 		/* skip if element is not in this file system */
347 		if (list_ent->rl_fsid != fscp->fs_cfsid)
348 			continue;
349 
350 		/* move from modified list to mf list */
351 		cachefs_rlent_moveto_nolock(cachep, CACHEFS_RL_MF, curp, 0);
352 	}
353 	mutex_exit(&cachep->c_contentslock);
354 }
355 
356 /*
357  * Moves the contents of the active list to the rl list.
358  * Leave modified files on the active list, so they are not
359  * garbage collected.
360  */
361 void
362 cachefs_rl_cleanup(cachefscache_t *cachep)
363 {
364 	cachefs_rl_listhead_t *lhp;
365 	rl_entry_t *rlp;
366 	uint_t entno, next;
367 	int error;
368 
369 	ASSERT(MUTEX_HELD(&cachep->c_contentslock));
370 
371 	/*
372 	 * if fsck ran, then both of these lists should be empty.  the
373 	 * only time this isn't the case is when we've done a cachefs
374 	 * boot with a clean cache.  then, the cache may have been
375 	 * clean, but files and attrfiles were left dangling.
376 	 *
377 	 * when this happens, we just fix the linked lists here.  this
378 	 * means that the attrcache header and cnode metadata might
379 	 * have incorrect information about which resource lists an
380 	 * entity is currently on.  so, we set CACHE_CHECK_RLTYPE,
381 	 * which says cache-wide to double-check and go with whatever
382 	 * is in the resource list at the time such an object is
383 	 * loaded into memory.
384 	 */
385 
386 	lhp = RL_HEAD(cachep, CACHEFS_RL_ACTIVE);
387 	if (lhp->rli_itemcnt > 0) {
388 		cachep->c_flags |= CACHE_CHECK_RLTYPE;
389 		cachefs_cache_dirty(cachep, 0);
390 	}
391 	for (entno = lhp->rli_front; entno != 0; entno = next) {
392 		error = cachefs_rl_entry_get(cachep, entno, &rlp);
393 		if (error)
394 			return;
395 		next = rlp->rl_bkwd_idx;
396 
397 		ASSERT(rlp->rl_current == CACHEFS_RL_ACTIVE);
398 		cachefs_rlent_moveto_nolock(cachep, CACHEFS_RL_GC, entno, 0);
399 	}
400 
401 #if 0
402 	lhp = RL_HEAD(cachep, CACHEFS_RL_ATTRFILE);
403 	if (lhp->rli_itemcnt > 0) {
404 		cachep->c_flags |= CACHE_CHECK_RLTYPE;
405 		cachefs_cache_dirty(cachep, 0);
406 	}
407 	for (entno = lhp->rli_front; entno != 0; entno = next) {
408 		error = cachefs_rl_entry_get(cachep, entno, &rlp);
409 		if (error)
410 			return;
411 		next = rlp->rl_bkwd_idx;
412 
413 		ASSERT(rlp->rl_current == CACHEFS_RL_ATTRFILE);
414 		cachefs_rlent_moveto_nolock(cachep, CACHEFS_RL_GC, entno, 0);
415 	}
416 #endif
417 }
418 
419 int
420 cachefs_allocfile(cachefscache_t *cachep)
421 {
422 	int error = 0;
423 	int collect = 0;
424 	struct statvfs64 sb;
425 	fsfilcnt64_t used;
426 
427 	(void) VFS_STATVFS(cachep->c_dirvp->v_vfsp, &sb);
428 	used = sb.f_files - sb.f_ffree;
429 
430 	mutex_enter(&cachep->c_contentslock);
431 
432 	/* if there are no more available inodes */
433 	if ((cachep->c_usage.cu_filesused >= cachep->c_label.cl_maxinodes) ||
434 	    ((cachep->c_usage.cu_filesused > cachep->c_label.cl_filemin) &&
435 	    (used > cachep->c_label.cl_filetresh))) {
436 		error = ENOSPC;
437 		if ((cachep->c_flags & CACHE_GARBAGE_COLLECT) == 0)
438 			collect = 1;
439 	}
440 
441 	/* else if there are more available inodes */
442 	else {
443 		cachefs_cache_dirty(cachep, 0);
444 		cachep->c_usage.cu_filesused++;
445 		if (((cachep->c_flags & CACHE_GARBAGE_COLLECT) == 0) &&
446 		    (cachep->c_usage.cu_filesused >=
447 		    cachep->c_label.cl_filehiwat))
448 			collect = 1;
449 	}
450 
451 	mutex_exit(&cachep->c_contentslock);
452 
453 	if (collect)
454 		cachefs_garbage_collect_queue(cachep);
455 
456 	return (error);
457 }
458 
459 void
460 cachefs_freefile(cachefscache_t *cachep)
461 {
462 	mutex_enter(&cachep->c_contentslock);
463 	ASSERT(cachep->c_usage.cu_filesused > 0);
464 	cachefs_cache_dirty(cachep, 0);
465 	cachep->c_usage.cu_filesused--;
466 	mutex_exit(&cachep->c_contentslock);
467 }
468 
469 /*ARGSUSED*/
470 int
471 cachefs_allocblocks(cachefscache_t *cachep, size_t nblks,
472     enum cachefs_rl_type type)
473 {
474 	int error = 0;
475 	int collect = 0;
476 	struct statvfs64 sb;
477 	size_t used;
478 	size_t blocks;
479 
480 	ASSERT(type != CACHEFS_RL_FREE);
481 
482 	(void) VFS_STATVFS(cachep->c_dirvp->v_vfsp, &sb);
483 	used = ((sb.f_blocks - sb.f_bfree) * sb.f_frsize) / MAXBSIZE;
484 
485 	mutex_enter(&cachep->c_contentslock);
486 
487 	/* if there are no more available blocks */
488 	blocks = cachep->c_usage.cu_blksused + nblks;
489 	if ((blocks >= cachep->c_label.cl_maxblks) ||
490 	    ((blocks > cachep->c_label.cl_blockmin) &&
491 	    (used > cachep->c_label.cl_blocktresh))) {
492 		error = ENOSPC;
493 		if ((cachep->c_flags & CACHE_GARBAGE_COLLECT) == 0)
494 			collect = 1;
495 	}
496 
497 	/* else if there are more available blocks */
498 	else {
499 		cachefs_cache_dirty(cachep, 0);
500 		cachep->c_usage.cu_blksused += (uint_t)nblks;
501 		ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
502 		RL_HEAD(cachep, type)->rli_blkcnt += nblks;
503 
504 		if (((cachep->c_flags & CACHE_GARBAGE_COLLECT) == 0) &&
505 		    (cachep->c_usage.cu_blksused >=
506 		    cachep->c_label.cl_blkhiwat))
507 			collect = 1;
508 	}
509 
510 	mutex_exit(&cachep->c_contentslock);
511 
512 	if (collect)
513 		cachefs_garbage_collect_queue(cachep);
514 
515 	return (error);
516 }
517 
518 void
519 cachefs_freeblocks(cachefscache_t *cachep, size_t nblks,
520 		enum cachefs_rl_type type)
521 {
522 	mutex_enter(&cachep->c_contentslock);
523 	cachefs_cache_dirty(cachep, 0);
524 	cachep->c_usage.cu_blksused -= (uint_t)nblks;
525 	ASSERT(cachep->c_usage.cu_blksused >= 0);
526 	ASSERT((CACHEFS_RL_START <= type) && (type <= CACHEFS_RL_END));
527 	ASSERT(type != CACHEFS_RL_FREE);
528 	RL_HEAD(cachep, type)->rli_blkcnt -= nblks;
529 	mutex_exit(&cachep->c_contentslock);
530 }
531 
532 int
533 cachefs_victim(cachefscache_t *cachep)
534 {
535 	uint_t entno;
536 	rl_entry_t *rl_ent;
537 	int error = 0;
538 	ino64_t fsid;
539 	cfs_cid_t cid;
540 	struct fscache *fscp;
541 	struct filegrp *fgp;
542 	struct cachefs_metadata md;
543 	struct cnode *cp;
544 	int isattrc;
545 	cachefs_rl_listhead_t *lhp;
546 
547 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
548 	fscp = NULL;
549 	fgp = NULL;
550 
551 	/* get the file and fsid of the first item on the rl list */
552 	/* XXX call rlent_data() instead */
553 	mutex_enter(&cachep->c_contentslock);
554 	lhp = RL_HEAD(cachep, CACHEFS_RL_GC);
555 	entno = lhp->rli_front;
556 	if (entno == 0) {
557 		mutex_exit(&cachep->c_contentslock);
558 		error = ENOSPC;
559 		goto out;
560 	}
561 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
562 	if (error) {
563 		mutex_exit(&cachep->c_contentslock);
564 		goto out;
565 	}
566 	fsid = rl_ent->rl_fsid;
567 	cid.cid_fileno = rl_ent->rl_fileno;
568 	ASSERT(rl_ent->rl_local == 0);
569 	cid.cid_flags = 0;
570 	isattrc = rl_ent->rl_attrc;
571 	mutex_exit(&cachep->c_contentslock);
572 
573 	/* get the file system cache object for this fsid */
574 	mutex_enter(&cachep->c_fslistlock);
575 	fscp = fscache_list_find(cachep, fsid);
576 	if (fscp == NULL) {
577 		fscp = fscache_create(cachep);
578 		error = fscache_activate(fscp, fsid, NULL, NULL, 0);
579 		if (error) {
580 			cmn_err(CE_WARN,
581 			    "cachefs: cache corruption, run fsck\n");
582 			fscache_destroy(fscp);
583 			fscp = NULL;
584 			mutex_exit(&cachep->c_fslistlock);
585 			error = 0;
586 			goto out;
587 		}
588 		fscache_list_add(cachep, fscp);
589 	}
590 	fscache_hold(fscp);
591 	mutex_exit(&cachep->c_fslistlock);
592 
593 	/* get the file group object for this file */
594 	mutex_enter(&fscp->fs_fslock);
595 	fgp = filegrp_list_find(fscp, &cid);
596 	if (fgp == NULL) {
597 		fgp = filegrp_create(fscp, &cid);
598 		filegrp_list_add(fscp, fgp);
599 	}
600 	if (fgp->fg_flags & CFS_FG_ALLOC_ATTR) {
601 		if (isattrc == 0) {
602 			cmn_err(CE_WARN,
603 			    "cachefs: cache corruption, run fsck\n");
604 			delay(5*hz);
605 		}
606 		filegrp_list_remove(fscp, fgp);
607 		filegrp_destroy(fgp);
608 		error = 0;
609 		fgp = NULL;
610 		mutex_exit(&fscp->fs_fslock);
611 		goto out;
612 	}
613 
614 	/* if we are victimizing an attrcache file */
615 	if (isattrc) {
616 		mutex_enter(&fgp->fg_mutex);
617 		/* if the filegrp is not writable */
618 		if ((fgp->fg_flags & CFS_FG_WRITE) == 0) {
619 			mutex_exit(&fgp->fg_mutex);
620 			error = EROFS;
621 			fgp = NULL;
622 			mutex_exit(&fscp->fs_fslock);
623 			goto out;
624 		}
625 
626 		/* if the filegrp did not go active on us */
627 		if ((fgp->fg_count == 0) && (fgp->fg_header->ach_nffs == 0)) {
628 			mutex_exit(&fgp->fg_mutex);
629 			filegrp_list_remove(fscp, fgp);
630 			fgp->fg_header->ach_count = 0;
631 			filegrp_destroy(fgp);
632 		} else {
633 #ifdef CFSDEBUG
634 			CFS_DEBUG(CFSDEBUG_RESOURCE)
635 				printf("c_victim: filegrp went active"
636 				    " %p %llu %d %d %lld\n",
637 				    (void *) fgp,
638 				    (u_longlong_t)fgp->fg_id.cid_fileno,
639 				    fgp->fg_header->ach_rlno,
640 				    fgp->fg_count, fgp->fg_header->ach_nffs);
641 #endif
642 			ASSERT(fgp->fg_header->ach_rl_current !=
643 			    CACHEFS_RL_GC);
644 			mutex_exit(&fgp->fg_mutex);
645 		}
646 		fgp = NULL;
647 		error = 0;
648 		mutex_exit(&fscp->fs_fslock);
649 		goto out;
650 	}
651 	ASSERT((fgp->fg_flags & CFS_FG_ALLOC_FILE) == 0);
652 	filegrp_hold(fgp);
653 	mutex_exit(&fscp->fs_fslock);
654 
655 	/* grab the cnode list lock */
656 	mutex_enter(&fgp->fg_cnodelock);
657 
658 	/* see if a cnode exists for this file */
659 	(void) cachefs_cnode_find(fgp, &cid, NULL, &cp, NULL, NULL);
660 	if (cp) {
661 		VN_HOLD(CTOV(cp));
662 
663 		/* move file from rl to active list */
664 		cachefs_rlent_moveto(fscp->fs_cache,
665 		    CACHEFS_RL_ACTIVE, cp->c_metadata.md_rlno,
666 		    cp->c_metadata.md_frontblks);
667 		cp->c_metadata.md_rltype = CACHEFS_RL_ACTIVE;
668 		mutex_exit(&cp->c_statelock);
669 		mutex_exit(&fgp->fg_cnodelock);
670 		VN_RELE(CTOV(cp));
671 		error = 0;
672 		goto out;
673 	}
674 
675 	/*
676 	 * The cnode does not exist and since we hold the hashlock
677 	 * it cannot be created until we are done.
678 	 */
679 
680 	/* see if the item is no longer on the rl list, it could happen */
681 	mutex_enter(&cachep->c_contentslock);
682 	lhp = RL_HEAD(cachep, CACHEFS_RL_GC);
683 	entno = lhp->rli_front;
684 	if (entno == 0) {
685 		mutex_exit(&cachep->c_contentslock);
686 		mutex_exit(&fgp->fg_cnodelock);
687 		error = ENOSPC;
688 		goto out;
689 	}
690 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
691 	if (error) {
692 		mutex_exit(&cachep->c_contentslock);
693 		mutex_exit(&fgp->fg_cnodelock);
694 		goto out;
695 	}
696 	if ((fsid != rl_ent->rl_fsid) ||
697 	    (cid.cid_fileno != rl_ent->rl_fileno)) {
698 		mutex_exit(&cachep->c_contentslock);
699 		mutex_exit(&fgp->fg_cnodelock);
700 		error = 0;
701 		goto out;
702 	}
703 	mutex_exit(&cachep->c_contentslock);
704 
705 	/* Get the metadata from the attrcache file */
706 	ASSERT((fgp->fg_flags & CFS_FG_ALLOC_ATTR) == 0);
707 	error = filegrp_read_metadata(fgp, &cid, &md);
708 	ASSERT(error == 0);
709 
710 	/* md.md_rltype may be incorrect, but we know file isn't active. */
711 	if (error) {
712 		/* XXX this should never happen, fix on panic */
713 		mutex_exit(&fgp->fg_cnodelock);
714 		error = 0;
715 		goto out;
716 	}
717 
718 	/* destroy the frontfile */
719 	cachefs_removefrontfile(&md, &cid, fgp);
720 
721 	/* remove the victim from the gc list */
722 	cachefs_rlent_moveto(fscp->fs_cache, CACHEFS_RL_FREE, entno, 0);
723 
724 	/* destroy the metadata */
725 	(void) filegrp_destroy_metadata(fgp, &cid);
726 
727 	mutex_exit(&fgp->fg_cnodelock);
728 	error = 0;
729 out:
730 	if (fgp) {
731 		filegrp_rele(fgp);
732 	}
733 	if (fscp) {
734 		fscache_rele(fscp);
735 	}
736 	return (error);
737 }
738 
739 static void
740 cachefs_garbage_collect(cachefscache_t *cachep)
741 {
742 	fsfilcnt64_t filelowat, filelowatmax, maxfiles, threshfiles;
743 	fsblkcnt64_t blocklowat, blocklowatmax, maxblks, threshblks;
744 	int error;
745 	struct cache_usage *cup = &cachep->c_usage;
746 
747 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
748 	mutex_enter(&cachep->c_contentslock);
749 	ASSERT(cachep->c_flags & CACHE_GARBAGE_COLLECT);
750 	filelowat = cachep->c_label.cl_filelowat;
751 	blocklowat = cachep->c_label.cl_blklowat;
752 	maxblks = cachep->c_label.cl_maxblks;
753 	maxfiles = cachep->c_label.cl_maxinodes;
754 	threshblks = cachep->c_label.cl_blocktresh;
755 	threshfiles = cachep->c_label.cl_filetresh;
756 	mutex_exit(&cachep->c_contentslock);
757 
758 	cachep->c_gc_count++;
759 	cachep->c_gc_time = time;
760 	cachep->c_gc_before = cachefs_gc_front_atime(cachep);
761 
762 	/*
763 	 * since we're here, we're running out of blocks or files.
764 	 * file and block lowat are what determine how low we garbage
765 	 * collect.  in order to do any good, we should drop below
766 	 * maxblocks, threshblocks, or the current blocks, whichever
767 	 * is smaller (same goes for files).  however, we won't go
768 	 * below an arbitrary (small) minimum for each.
769 	 */
770 
771 	/* move down for maxfiles and maxblocks */
772 	if ((filelowatmax = (maxfiles * 7) / 10) < filelowat)
773 		filelowat = filelowatmax;
774 	if ((blocklowatmax = (maxblks * 7) / 10) < blocklowat)
775 		blocklowat = blocklowatmax;
776 
777 	/* move down for threshfiles and threshblocks */
778 	if ((filelowatmax = (threshfiles * 7) / 10) < filelowat)
779 		filelowat = filelowatmax;
780 	if ((blocklowatmax = (threshblks * 7) / 10) < blocklowat)
781 		blocklowat = blocklowatmax;
782 
783 	/* move down for current files and blocks */
784 	if ((filelowatmax = ((fsfilcnt64_t)cup->cu_filesused * 7) / 10) <
785 	    filelowat)
786 		filelowat = filelowatmax;
787 	if ((blocklowatmax = ((fsblkcnt64_t)cup->cu_blksused * 7) / 10) <
788 	    blocklowat)
789 		blocklowat = blocklowatmax;
790 
791 	/* move up for an arbitrary minimum */
792 #define	MIN_BLKLO	640		/* 640*8192 == 5MB */
793 #define	MIN_FILELO	1000
794 	if (filelowat < MIN_FILELO)
795 		filelowat = MIN_FILELO;
796 	if (blocklowat < MIN_BLKLO)
797 		blocklowat = MIN_BLKLO;
798 
799 	while (cup->cu_filesused > filelowat || cup->cu_blksused > blocklowat) {
800 		/* if the thread is to terminate */
801 		if (cachep->c_flags & CACHE_CACHEW_THREADEXIT)
802 			break;
803 
804 		error = cachefs_victim(cachep);
805 		if (error)
806 			break;
807 	}
808 
809 	cachep->c_gc_after = cachefs_gc_front_atime(cachep);
810 	CACHEFS_TIME_TO_CFS_TIME_COPY(cachep->c_gc_after,
811 			cachep->c_rlinfo.rl_gctime, error);
812 }
813 
814 /*
815  * traverse the packed pending list, repacking files when possible.
816  */
817 
818 static void
819 cachefs_packed_pending(cachefscache_t *cachep)
820 {
821 	rl_entry_t rl;
822 	int error = 0; /* not returned -- used as placeholder */
823 	fscache_t *fscp = NULL;
824 	cfs_cid_t cid;
825 	cnode_t *cp;
826 	uint_t entno;
827 	int count = 0;
828 	cachefs_rl_listhead_t *lhp;
829 
830 	ASSERT(MUTEX_HELD(&cachep->c_contentslock));
831 
832 	lhp = RL_HEAD(cachep, CACHEFS_RL_PACKED_PENDING);
833 	count = lhp->rli_itemcnt;
834 
835 	mutex_exit(&cachep->c_contentslock);
836 
837 	rl.rl_current = CACHEFS_RL_PACKED_PENDING;
838 	while (cachefs_rlent_data(cachep, &rl, &entno) == 0) {
839 		if (count-- <= 0) {
840 #ifdef CFSDEBUG
841 			CFS_DEBUG(CFSDEBUG_RESOURCE)
842 				printf("cachefs_ppending: count exceeded\n");
843 #endif /* CFSDEBUG */
844 			break;
845 		}
846 		if ((cachep->c_flags &
847 		    (CACHE_PACKED_PENDING | CACHE_CACHEW_THREADEXIT)) !=
848 		    CACHE_PACKED_PENDING) {
849 #ifdef CFSDEBUG
850 			CFS_DEBUG(CFSDEBUG_RESOURCE)
851 				printf("cachefs_ppending: early exit\n");
852 #endif /* CFSDEBUG */
853 			break;
854 		}
855 		if (rl.rl_current != CACHEFS_RL_PACKED_PENDING) {
856 #ifdef CFSDEBUG
857 			CFS_DEBUG(CFSDEBUG_RESOURCE)
858 				printf("cachefs_ppending: gone from list\n");
859 #endif /* CFSDEBUG */
860 			break;
861 		}
862 
863 		/* if the fscp we have does not match */
864 		if ((fscp == NULL) || (fscp->fs_cfsid != rl.rl_fsid)) {
865 			if (fscp) {
866 				cachefs_cd_release(fscp);
867 				fscache_rele(fscp);
868 				fscp = NULL;
869 			}
870 
871 			/* get the file system cache object for this fsid */
872 			mutex_enter(&cachep->c_fslistlock);
873 			fscp = fscache_list_find(cachep, rl.rl_fsid);
874 			if (fscp == NULL) {
875 
876 				/*
877 				 * uh oh, the filesystem probably
878 				 * isn't mounted.  we `move' this
879 				 * entry onto the same list that it's
880 				 * on, which really just moves it to
881 				 * the back of the list.  we need not
882 				 * worry about an infinite loop, due
883 				 * to the counter.
884 				 */
885 
886 				cachefs_rlent_moveto(cachep,
887 				    CACHEFS_RL_PACKED_PENDING, entno, 0);
888 #ifdef CFSDEBUG
889 				CFS_DEBUG(CFSDEBUG_RESOURCE)
890 					printf("cachefs_ppending: "
891 					    "fscp find failed\n");
892 #endif /* CFSDEBUG */
893 				continue;
894 			}
895 			fscache_hold(fscp);
896 			mutex_exit(&cachep->c_fslistlock);
897 
898 			/* get access to the file system */
899 			error = cachefs_cd_access(fscp, 0, 0);
900 			if ((error) ||
901 			    (fscp->fs_cdconnected != CFS_CD_CONNECTED)) {
902 #ifdef CFSDEBUG
903 				CFS_DEBUG(CFSDEBUG_RESOURCE)
904 					printf("cachefs: "
905 					    "ppending: err %d con %d\n",
906 					    error, fscp->fs_cdconnected);
907 #endif /* CFSDEBUG */
908 				fscache_rele(fscp);
909 				fscp = NULL;
910 				break;
911 			}
912 		}
913 
914 		/* get the cnode for the file */
915 		cid.cid_fileno = rl.rl_fileno;
916 		cid.cid_flags = rl.rl_local ? CFS_CID_LOCAL : 0;
917 		error = cachefs_cnode_make(&cid, fscp,
918 		    NULL, NULL, NULL, kcred, 0, &cp);
919 		if (error) {
920 #ifdef CFSDEBUG
921 			CFS_DEBUG(CFSDEBUG_RESOURCE)
922 				printf("cachefs: "
923 				    "ppending: could not find %llu\n",
924 				    (u_longlong_t)cid.cid_fileno);
925 			delay(5*hz);
926 #endif /* CFSDEBUG */
927 			break;
928 		}
929 
930 		mutex_enter(&cp->c_statelock);
931 		if (cp->c_flags & CN_STALE) {
932 			/* back file went away behind our back */
933 			ASSERT(cp->c_metadata.md_rlno == 0);
934 			mutex_exit(&cp->c_statelock);
935 
936 #ifdef CFSDEBUG
937 			CFS_DEBUG(CFSDEBUG_RESOURCE)
938 				printf("cachefs: ppending: stale\n");
939 #endif /* CFSDEBUG */
940 
941 			VN_RELE(CTOV(cp));
942 			continue;
943 		}
944 		mutex_exit(&cp->c_statelock);
945 
946 		error = cachefs_pack_common(CTOV(cp),
947 		    (cp->c_cred) ? cp->c_cred : kcred);
948 		VN_RELE(CTOV(cp));
949 
950 		if (error != 0) {
951 #ifdef CFSDEBUG
952 			CFS_DEBUG(CFSDEBUG_RESOURCE)
953 				printf("cachefs: "
954 				    "ppending: pack_common: error = %d\n",
955 				    error);
956 #endif /* CFSDEBUG */
957 			break;
958 		}
959 	}
960 
961 	if (fscp != NULL) {
962 		cachefs_cd_release(fscp);
963 		fscache_rele(fscp);
964 	}
965 
966 	mutex_enter(&cachep->c_contentslock);
967 	if (lhp->rli_itemcnt == 0)
968 		cachep->c_flags &= ~CACHE_PACKED_PENDING;
969 }
970 
971 /* seconds; interval to do ppend list */
972 static time_t cachefs_ppend_time = 900;
973 
974 /* main routine for the cachep worker thread */
975 void
976 cachefs_cachep_worker_thread(cachefscache_t *cachep)
977 {
978 	int error;
979 	struct flock64 fl;
980 	callb_cpr_t cprinfo;
981 	kmutex_t cpr_lock;
982 
983 
984 	/* lock the lock file for exclusive write access */
985 	fl.l_type = F_WRLCK;
986 	fl.l_whence = 0;
987 	fl.l_start = (offset_t)0;
988 	fl.l_len = (offset_t)1024;
989 	fl.l_sysid = 0;
990 	fl.l_pid = 0;
991 	error = VOP_FRLOCK(cachep->c_lockvp, F_SETLK, &fl, FWRITE,
992 		(offset_t)0, NULL, kcred);
993 	if (error) {
994 		cmn_err(CE_WARN,
995 		    "cachefs: Can't lock Cache Lock File(r); Error %d\n",
996 		    error);
997 	}
998 
999 	mutex_init(&cpr_lock, NULL, MUTEX_DEFAULT, NULL);
1000 	CALLB_CPR_INIT(&cprinfo, &cpr_lock, callb_generic_cpr, "cfs_gct");
1001 	mutex_enter(&cpr_lock);
1002 	mutex_enter(&cachep->c_contentslock);
1003 
1004 	/* loop while the thread is allowed to run */
1005 	while ((cachep->c_flags & CACHE_CACHEW_THREADEXIT) == 0) {
1006 		clock_t wakeup;
1007 
1008 		/* wait for a wakeup call */
1009 		cachep->c_flags &= ~CACHE_GARBAGE_COLLECT;
1010 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
1011 		mutex_exit(&cpr_lock);
1012 		wakeup = (clock_t)(lbolt + (cachefs_ppend_time * hz));
1013 		(void) cv_timedwait(&cachep->c_cwcv,
1014 			&cachep->c_contentslock, wakeup);
1015 		mutex_enter(&cpr_lock);
1016 		CALLB_CPR_SAFE_END(&cprinfo, &cpr_lock);
1017 
1018 		/* if the thread is to terminate */
1019 		if (cachep->c_flags & CACHE_CACHEW_THREADEXIT)
1020 			break;
1021 
1022 		/* thread is running during nofill, but just to hold lock */
1023 		if (cachep->c_flags & CACHE_NOFILL)
1024 			continue;
1025 
1026 		/* if garbage collection is to run */
1027 		if (cachep->c_flags & CACHE_GARBAGE_COLLECT) {
1028 			mutex_exit(&cachep->c_contentslock);
1029 			cachefs_garbage_collect(cachep);
1030 
1031 			/*
1032 			 * Prevent garbage collection from running more
1033 			 * than once every 30 seconds.  This addresses
1034 			 * those cases which do not allow removing
1035 			 * an item from the rl by keeping gc from
1036 			 * being a spin loop.
1037 			 */
1038 			delay(30*hz); /* XXX sam: still do this? */
1039 			mutex_enter(&cachep->c_contentslock);
1040 		}
1041 
1042 		if (cachep->c_flags & CACHE_PACKED_PENDING)
1043 			cachefs_packed_pending(cachep);
1044 		ASSERT(MUTEX_HELD(&cachep->c_contentslock));
1045 	}
1046 
1047 	cachep->c_flags &= ~CACHE_CACHEW_THREADRUN;
1048 	cv_broadcast(&cachep->c_cwhaltcv);
1049 	CALLB_CPR_EXIT(&cprinfo);
1050 	mutex_exit(&cachep->c_contentslock);
1051 	mutex_destroy(&cpr_lock);
1052 
1053 	/* unlock the lock file */
1054 	fl.l_type = F_UNLCK;
1055 	fl.l_whence = 0;
1056 	fl.l_start = (offset_t)0;
1057 	fl.l_len = (offset_t)1024;
1058 	fl.l_sysid = 0;
1059 	fl.l_pid = 0;
1060 	error = VOP_FRLOCK(cachep->c_lockvp, F_SETLK, &fl,
1061 		FWRITE, (offset_t)0, NULL, kcred);
1062 	if (error) {
1063 		cmn_err(CE_WARN, "cachefs: Can't unlock lock file\n");
1064 	}
1065 
1066 	thread_exit();
1067 	/*NOTREACHED*/
1068 }
1069 
1070 /* queues up a request to run the garbage collection */
1071 void
1072 cachefs_garbage_collect_queue(cachefscache_t *cachep)
1073 {
1074 	cachefs_rl_listhead_t *lhp;
1075 
1076 	ASSERT((cachep->c_flags & (CACHE_NOCACHE|CACHE_NOFILL)) == 0);
1077 	mutex_enter(&cachep->c_contentslock);
1078 
1079 	/* quit if there is no garbage collection thread */
1080 	if ((cachep->c_flags & CACHE_CACHEW_THREADRUN) == 0) {
1081 		mutex_exit(&cachep->c_contentslock);
1082 		return;
1083 	}
1084 
1085 	/* quit if garbage collection is already in progress */
1086 	if (cachep->c_flags & CACHE_GARBAGE_COLLECT) {
1087 		mutex_exit(&cachep->c_contentslock);
1088 		return;
1089 	}
1090 
1091 	/* quit if there is no garbage to collect */
1092 	lhp = RL_HEAD(cachep, CACHEFS_RL_GC);
1093 	if (lhp->rli_front == 0) {
1094 		mutex_exit(&cachep->c_contentslock);
1095 		return;
1096 	}
1097 
1098 	/* indicate garbage collecting is in progress */
1099 	cachep->c_flags |= CACHE_GARBAGE_COLLECT;
1100 
1101 	/* wake up the garbage collection thread */
1102 	cv_signal(&cachep->c_cwcv);
1103 
1104 	mutex_exit(&cachep->c_contentslock);
1105 }
1106 
1107 #ifdef CFSRLDEBUG
1108 time_t cachefs_dbvalid = 123; /* default to non-zero junk */
1109 struct kmem_cache *cachefs_rl_debug_cache = NULL;
1110 static int cachefs_rl_debug_maxcount = CACHEFS_RLDB_DEF_MAXCOUNT;
1111 kmutex_t cachefs_rl_debug_mutex;
1112 static int cachefs_rl_debug_inuse = 0;
1113 
1114 void
1115 cachefs_rl_debug_reclaim(void *cdrarg)
1116 {
1117 	extern cachefscache_t *cachefs_cachelist;
1118 	cachefscache_t *cachep;
1119 	int index;
1120 	int error;
1121 
1122 	for (cachep = cachefs_cachelist; cachep != NULL;
1123 		cachep = cachep->c_next) {
1124 		mutex_enter(&cachep->c_contentslock);
1125 
1126 		for (index = 0;
1127 		    index <= cachep->c_rlinfo.rl_entries;
1128 		    index++) {
1129 			rl_entry_t *rlent;
1130 
1131 			error = cachefs_rl_entry_get(cachep, index, &rlent);
1132 			if (error)
1133 				break;
1134 			cachefs_rl_debug_destroy(rlent);
1135 		}
1136 
1137 		mutex_exit(&cachep->c_contentslock);
1138 	}
1139 }
1140 
1141 void
1142 cachefs_rl_debug_save(rl_entry_t *rlent)
1143 {
1144 	rl_debug_t *rldb, *prev, *next;
1145 	int count = 0;
1146 
1147 	mutex_enter(&cachefs_rl_debug_mutex);
1148 	if (cachefs_rl_debug_cache == NULL)
1149 		cachefs_rl_debug_cache =
1150 		    kmem_cache_create("cachefs_rl_debug",
1151 			sizeof (rl_debug_t), 0,
1152 			NULL, NULL, cachefs_rl_debug_reclaim, NULL, NULL, 0);
1153 
1154 	rldb = kmem_cache_alloc(cachefs_rl_debug_cache, KM_SLEEP);
1155 	++cachefs_rl_debug_inuse;
1156 
1157 	rldb->db_hrtime = gethrtime();
1158 
1159 	rldb->db_attrc = rlent->rl_attrc;
1160 	rldb->db_fsck = rlent->rl_fsck;
1161 	rldb->db_fsid = rlent->rl_fsid;
1162 	rldb->db_fileno = rlent->rl_fileno;
1163 	rldb->db_current = rlent->rl_current;
1164 
1165 	rldb->db_stackheight = getpcstack(rldb->db_stack,
1166 	    CACHEFS_RLDB_STACKSIZE);
1167 
1168 	if (rlent->rl_dbvalid == cachefs_dbvalid) {
1169 		rldb->db_next = rlent->rl_debug;
1170 	} else {
1171 		rldb->db_next = NULL;
1172 		rlent->rl_dbvalid = cachefs_dbvalid;
1173 	}
1174 	rlent->rl_debug = rldb;
1175 
1176 	prev = rldb;
1177 	for (rldb = rldb->db_next; rldb != NULL; rldb = next) {
1178 		next = rldb->db_next;
1179 		if (++count >= cachefs_rl_debug_maxcount) {
1180 			if (prev != NULL)
1181 				prev->db_next = NULL;
1182 			kmem_cache_free(cachefs_rl_debug_cache, rldb);
1183 			--cachefs_rl_debug_inuse;
1184 			prev = NULL;
1185 		} else {
1186 			prev = rldb;
1187 		}
1188 	}
1189 	mutex_exit(&cachefs_rl_debug_mutex);
1190 }
1191 
1192 void
1193 cachefs_rl_debug_show(rl_entry_t *rlent)
1194 {
1195 	rl_debug_t *rldb;
1196 	hrtime_t now, elapse;
1197 	timestruc_t tv;
1198 	char *cname = NULL;
1199 	int i;
1200 
1201 	mutex_enter(&cachefs_rl_debug_mutex);
1202 	if (rlent->rl_dbvalid != cachefs_dbvalid) {
1203 		printf("cachefs_rldb: rl entry at %lx -- no info!\n",
1204 		    (uintptr_t)rlent);
1205 		mutex_exit(&cachefs_rl_debug_mutex);
1206 		return;
1207 	}
1208 
1209 	now = gethrtime();
1210 	hrt2ts(now, &tv);
1211 
1212 	printf("===== cachefs_rldb start at %ld =====\n", tv.tv_sec);
1213 	printf("-==== i am thread id %lx   ====-\n", (uintptr_t)curthread);
1214 
1215 	for (rldb = rlent->rl_debug;
1216 	    rldb != NULL;
1217 	    rldb = rldb->db_next) {
1218 		printf("----- cachefs_rldb record start -----\n");
1219 		elapse = now - rldb->db_hrtime;
1220 		hrt2ts(elapse, &tv);
1221 		printf("cachefs_rldb: ago = %lds %ldus\n",
1222 		    tv.tv_sec, tv.tv_nsec / 1000);
1223 
1224 		printf("cachefs_rldb: rl_attrc = %d\n", rldb->db_attrc);
1225 		printf("cachefs_rldb: rl_fsck = %d\n", rldb->db_fsck);
1226 		printf("cachefs_rldb: rl_fsid = %u\n", rldb->db_fsid);
1227 		printf("cachefs_rldb: rl_fileno = %lu\n", rldb->db_fileno);
1228 
1229 		switch (rldb->db_current) {
1230 		case CACHEFS_RL_NONE:
1231 			cname = "CACHEFS_RL_NONE";
1232 			break;
1233 		case CACHEFS_RL_FREE:
1234 			cname = "CACHEFS_RL_FREE";
1235 			break;
1236 		case CACHEFS_RL_GC:
1237 			cname = "CACHEFS_RL_GC";
1238 			break;
1239 		case CACHEFS_RL_ACTIVE:
1240 			cname = "CACHEFS_RL_ACTIVE";
1241 			break;
1242 		case CACHEFS_RL_ATTRFILE:
1243 			cname = "CACHEFS_RL_ATTRFILE";
1244 			break;
1245 		case CACHEFS_RL_MODIFIED:
1246 			cname = "CACHEFS_RL_MODIFIED";
1247 			break;
1248 		case CACHEFS_RL_PACKED:
1249 			cname = "CACHEFS_RL_PACKED";
1250 			break;
1251 		case CACHEFS_RL_PACKED_PENDING:
1252 			cname = "CACHEFS_RL_PACKED_PENDING";
1253 			break;
1254 		case CACHEFS_RL_MF:
1255 			cname = "CACHEFS_MF_GC";
1256 			break;
1257 		}
1258 		if (cname != NULL) {
1259 			printf("cachefs_rldb: state = %s\n", cname);
1260 		} else {
1261 			printf("cachefs_rldb: undefined state %x\n",
1262 			    rldb->db_current);
1263 		}
1264 
1265 		printf("cachefs_rldb: stack trace\n");
1266 		for (i = 0; i < rldb->db_stackheight; i++) {
1267 			char *sym;
1268 			uint_t off;
1269 
1270 			sym = kobj_getsymname(rldb->db_stack[i], &off);
1271 			printf("cachefs_rldb:    %s+%lx\n",
1272 			    sym ? sym : "?", off);
1273 			delay(hz/4);
1274 		}
1275 
1276 		printf("----- cachefs_rldb record end -----\n");
1277 	}
1278 
1279 	mutex_exit(&cachefs_rl_debug_mutex);
1280 }
1281 
1282 void
1283 cachefs_rl_debug_destroy(rl_entry_t *rlent)
1284 {
1285 	rl_debug_t *rldb, *next;
1286 
1287 	mutex_enter(&cachefs_rl_debug_mutex);
1288 	if (rlent->rl_dbvalid != cachefs_dbvalid) {
1289 		rlent->rl_debug = NULL;
1290 		mutex_exit(&cachefs_rl_debug_mutex);
1291 		return;
1292 	}
1293 
1294 	for (rldb = rlent->rl_debug; rldb != NULL; rldb = next) {
1295 		next = rldb->db_next;
1296 		kmem_cache_free(cachefs_rl_debug_cache, rldb);
1297 		--cachefs_rl_debug_inuse;
1298 	}
1299 
1300 	rlent->rl_debug = NULL;
1301 	mutex_exit(&cachefs_rl_debug_mutex);
1302 }
1303 #endif /* CFSRLDEBUG */
1304 
1305 int
1306 cachefs_rl_entry_get(cachefscache_t *cachep, uint_t entno, rl_entry_t **ent)
1307 {
1308 	rl_entry_t *rl_ent;
1309 	uint_t whichwindow, winoffset;
1310 	int error = 0;
1311 
1312 	ASSERT(MUTEX_HELD(&cachep->c_contentslock));
1313 	ASSERT(entno <= cachep->c_label.cl_maxinodes); /* strictly less? */
1314 #if 0
1315 	ASSERT((cachep->c_flags & CACHE_NOFILL) == 0);
1316 #endif
1317 
1318 	whichwindow = entno / CACHEFS_RLPMBS;
1319 	winoffset = entno % CACHEFS_RLPMBS;
1320 
1321 	if ((cachep->c_rl_entries == NULL) ||
1322 	    (cachep->c_rl_window != whichwindow)) {
1323 		if (cachep->c_rl_entries != NULL) {
1324 			error = vn_rdwr(UIO_WRITE, cachep->c_resfilevp,
1325 			    (caddr_t)cachep->c_rl_entries, MAXBSIZE,
1326 			    (offset_t)((cachep->c_rl_window + 1) * MAXBSIZE),
1327 			    UIO_SYSSPACE, 0, RLIM_INFINITY, kcred, NULL);
1328 			if (error)
1329 				return (error);
1330 		}
1331 		else
1332 			cachep->c_rl_entries = (rl_entry_t *)
1333 			    cachefs_kmem_alloc(MAXBSIZE, KM_SLEEP);
1334 
1335 		error = vn_rdwr(UIO_READ, cachep->c_resfilevp,
1336 		    (caddr_t)cachep->c_rl_entries, MAXBSIZE,
1337 		    (offset_t)((whichwindow + 1) * MAXBSIZE),
1338 		    UIO_SYSSPACE, 0, RLIM_INFINITY, kcred, NULL);
1339 		if (error) {
1340 			cachefs_kmem_free(cachep->c_rl_entries, MAXBSIZE);
1341 			cachep->c_rl_entries = NULL;
1342 			return (error);
1343 		}
1344 		cachep->c_rl_window = whichwindow;
1345 	}
1346 	rl_ent = &cachep->c_rl_entries[winoffset];
1347 
1348 	*ent = rl_ent;
1349 #ifdef CFSRLDEBUG
1350 	cachefs_rl_debug_save(rl_ent);
1351 #endif /* CFSRLDEBUG */
1352 
1353 	return (error);
1354 }
1355 
1356 static time_t
1357 cachefs_gc_front_atime(cachefscache_t *cachep)
1358 {
1359 	char namebuf[CFS_FRONTFILE_NAME_SIZE];
1360 
1361 	rl_entry_t rl, *rl_ent;
1362 	uint_t entno, fgsize;
1363 	cfs_cid_t dircid, cid;
1364 	struct fscache *fscp;
1365 	cachefs_rl_listhead_t *lhp;
1366 	int error;
1367 
1368 	struct vnode *dirvp, *filevp;
1369 	struct vattr va;
1370 
1371 	int reledir = 0;
1372 	int gotfile = 0;
1373 	time_t rc = (time_t)0;
1374 
1375 	mutex_enter(&cachep->c_contentslock);
1376 	lhp = RL_HEAD(cachep, CACHEFS_RL_GC);
1377 	entno = lhp->rli_front;
1378 	if (entno == 0) {
1379 		mutex_exit(&cachep->c_contentslock);
1380 		goto out;
1381 	}
1382 
1383 	error = cachefs_rl_entry_get(cachep, entno, &rl_ent);
1384 	if (error) {
1385 		mutex_exit(&cachep->c_contentslock);
1386 		goto out;
1387 	}
1388 	rl = *rl_ent;
1389 	mutex_exit(&cachep->c_contentslock);
1390 	cid.cid_fileno = rl.rl_fileno;
1391 	ASSERT(rl.rl_local == 0);
1392 	cid.cid_flags = 0;
1393 	dircid.cid_flags = 0;
1394 	mutex_enter(&cachep->c_fslistlock);
1395 	if ((fscp = fscache_list_find(cachep, rl.rl_fsid)) == NULL) {
1396 		mutex_exit(&cachep->c_fslistlock);
1397 		goto out;
1398 	}
1399 
1400 	if (rl.rl_attrc) {
1401 		make_ascii_name(&cid, namebuf);
1402 		dirvp = fscp->fs_fsattrdir;
1403 	} else {
1404 		dirvp = NULL;
1405 		fgsize = fscp->fs_info.fi_fgsize;
1406 		dircid.cid_fileno = ((cid.cid_fileno / fgsize) * fgsize);
1407 		make_ascii_name(&dircid, namebuf);
1408 		if (VOP_LOOKUP(fscp->fs_fscdirvp, namebuf,
1409 		    &dirvp, (struct pathname *)NULL, 0,
1410 		    (vnode_t *)NULL, kcred) == 0) {
1411 			make_ascii_name(&cid, namebuf);
1412 			reledir++;
1413 		} else {
1414 			mutex_exit(&cachep->c_fslistlock);
1415 			goto out;
1416 		}
1417 	}
1418 	if (dirvp && VOP_LOOKUP(dirvp, namebuf, &filevp,
1419 	    (struct pathname *)NULL, 0,
1420 	    (vnode_t *)NULL, kcred) == 0) {
1421 		gotfile = 1;
1422 	}
1423 	if (reledir)
1424 		VN_RELE(dirvp);
1425 	mutex_exit(&cachep->c_fslistlock);
1426 
1427 	if (gotfile) {
1428 		va.va_mask = AT_ATIME;
1429 		if (VOP_GETATTR(filevp, &va, 0, kcred) == 0)
1430 			rc = va.va_atime.tv_sec;
1431 		VN_RELE(filevp);
1432 	}
1433 
1434 out:
1435 	return (rc);
1436 }
1437