pseudofs_vncache.c revision 175294 
1/*-
2 * Copyright (c) 2001 Dag-Erling Co�dan Sm�rgrav
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer
10 *    in this position and unchanged.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote products
15 *    derived from this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: head/sys/fs/pseudofs/pseudofs_vncache.c 175294 2008-01-13 14:44:15Z attilio $");
31
32#include "opt_pseudofs.h"
33
34#include <sys/param.h>
35#include <sys/kernel.h>
36#include <sys/systm.h>
37#include <sys/eventhandler.h>
38#include <sys/lock.h>
39#include <sys/malloc.h>
40#include <sys/mutex.h>
41#include <sys/proc.h>
42#include <sys/sysctl.h>
43#include <sys/vnode.h>
44
45#include <fs/pseudofs/pseudofs.h>
46#include <fs/pseudofs/pseudofs_internal.h>
47
48static MALLOC_DEFINE(M_PFSVNCACHE, "pfs_vncache", "pseudofs vnode cache");
49
50static struct mtx pfs_vncache_mutex;
51static struct pfs_vdata *pfs_vncache;
52static eventhandler_tag pfs_exit_tag;
53static void pfs_exit(void *arg, struct proc *p);
54
55SYSCTL_NODE(_vfs_pfs, OID_AUTO, vncache, CTLFLAG_RW, 0,
56    "pseudofs vnode cache");
57
58static int pfs_vncache_entries;
59SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, entries, CTLFLAG_RD,
60    &pfs_vncache_entries, 0,
61    "number of entries in the vnode cache");
62
63static int pfs_vncache_maxentries;
64SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, maxentries, CTLFLAG_RD,
65    &pfs_vncache_maxentries, 0,
66    "highest number of entries in the vnode cache");
67
68static int pfs_vncache_hits;
69SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, hits, CTLFLAG_RD,
70    &pfs_vncache_hits, 0,
71    "number of cache hits since initialization");
72
73static int pfs_vncache_misses;
74SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, misses, CTLFLAG_RD,
75    &pfs_vncache_misses, 0,
76    "number of cache misses since initialization");
77
78extern struct vop_vector pfs_vnodeops;	/* XXX -> .h file */
79
80/*
81 * Initialize vnode cache
82 */
83void
84pfs_vncache_load(void)
85{
86
87	mtx_assert(&Giant, MA_OWNED);
88	mtx_init(&pfs_vncache_mutex, "pfs_vncache", NULL, MTX_DEF);
89	pfs_exit_tag = EVENTHANDLER_REGISTER(process_exit, pfs_exit, NULL,
90	    EVENTHANDLER_PRI_ANY);
91}
92
93/*
94 * Tear down vnode cache
95 */
96void
97pfs_vncache_unload(void)
98{
99
100	mtx_assert(&Giant, MA_OWNED);
101	EVENTHANDLER_DEREGISTER(process_exit, pfs_exit_tag);
102	KASSERT(pfs_vncache_entries == 0,
103	    ("%d vncache entries remaining", pfs_vncache_entries));
104	mtx_destroy(&pfs_vncache_mutex);
105}
106
107/*
108 * Allocate a vnode
109 */
110int
111pfs_vncache_alloc(struct mount *mp, struct vnode **vpp,
112		  struct pfs_node *pn, pid_t pid)
113{
114	struct pfs_vdata *pvd;
115	struct vnode *vp;
116	int error;
117
118	/*
119	 * See if the vnode is in the cache.
120	 * XXX linear search is not very efficient.
121	 */
122retry:
123	mtx_lock(&pfs_vncache_mutex);
124	for (pvd = pfs_vncache; pvd; pvd = pvd->pvd_next) {
125		if (pvd->pvd_pn == pn && pvd->pvd_pid == pid &&
126		    pvd->pvd_vnode->v_mount == mp) {
127			vp = pvd->pvd_vnode;
128			VI_LOCK(vp);
129			mtx_unlock(&pfs_vncache_mutex);
130			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
131				++pfs_vncache_hits;
132				*vpp = vp;
133				/*
134				 * Some callers cache_enter(vp) later, so
135				 * we have to make sure it's not in the
136				 * VFS cache so it doesn't get entered
137				 * twice.  A better solution would be to
138				 * make pfs_vncache_alloc() responsible
139				 * for entering the vnode in the VFS
140				 * cache.
141				 */
142				cache_purge(vp);
143				return (0);
144			}
145			goto retry;
146		}
147	}
148	mtx_unlock(&pfs_vncache_mutex);
149	++pfs_vncache_misses;
150
151	/* nope, get a new one */
152	MALLOC(pvd, struct pfs_vdata *, sizeof *pvd, M_PFSVNCACHE, M_WAITOK);
153	mtx_lock(&pfs_vncache_mutex);
154	if (++pfs_vncache_entries > pfs_vncache_maxentries)
155		pfs_vncache_maxentries = pfs_vncache_entries;
156	mtx_unlock(&pfs_vncache_mutex);
157	error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp);
158	if (error) {
159		mtx_lock(&pfs_vncache_mutex);
160		--pfs_vncache_entries;
161		mtx_unlock(&pfs_vncache_mutex);
162		FREE(pvd, M_PFSVNCACHE);
163		return (error);
164	}
165	pvd->pvd_pn = pn;
166	pvd->pvd_pid = pid;
167	(*vpp)->v_data = pvd;
168	switch (pn->pn_type) {
169	case pfstype_root:
170		(*vpp)->v_vflag = VV_ROOT;
171#if 0
172		printf("root vnode allocated\n");
173#endif
174		/* fall through */
175	case pfstype_dir:
176	case pfstype_this:
177	case pfstype_parent:
178	case pfstype_procdir:
179		(*vpp)->v_type = VDIR;
180		break;
181	case pfstype_file:
182		(*vpp)->v_type = VREG;
183		break;
184	case pfstype_symlink:
185		(*vpp)->v_type = VLNK;
186		break;
187	case pfstype_none:
188		KASSERT(0, ("pfs_vncache_alloc called for null node\n"));
189	default:
190		panic("%s has unexpected type: %d", pn->pn_name, pn->pn_type);
191	}
192	/*
193	 * Propagate flag through to vnode so users know it can change
194	 * if the process changes (i.e. execve)
195	 */
196	if ((pn->pn_flags & PFS_PROCDEP) != 0)
197		(*vpp)->v_vflag |= VV_PROCDEP;
198	pvd->pvd_vnode = *vpp;
199	(*vpp)->v_vnlock->lk_flags |= LK_CANRECURSE;
200	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
201	error = insmntque(*vpp, mp);
202	if (error != 0) {
203		mtx_lock(&pfs_vncache_mutex);
204		--pfs_vncache_entries;
205		mtx_unlock(&pfs_vncache_mutex);
206		FREE(pvd, M_PFSVNCACHE);
207		*vpp = NULLVP;
208		return (error);
209	}
210	mtx_lock(&pfs_vncache_mutex);
211	pvd->pvd_prev = NULL;
212	pvd->pvd_next = pfs_vncache;
213	if (pvd->pvd_next)
214		pvd->pvd_next->pvd_prev = pvd;
215	pfs_vncache = pvd;
216	mtx_unlock(&pfs_vncache_mutex);
217	return (0);
218}
219
220/*
221 * Free a vnode
222 */
223int
224pfs_vncache_free(struct vnode *vp)
225{
226	struct pfs_vdata *pvd;
227
228	mtx_lock(&pfs_vncache_mutex);
229	pvd = (struct pfs_vdata *)vp->v_data;
230	KASSERT(pvd != NULL, ("pfs_vncache_free(): no vnode data\n"));
231	if (pvd->pvd_next)
232		pvd->pvd_next->pvd_prev = pvd->pvd_prev;
233	if (pvd->pvd_prev)
234		pvd->pvd_prev->pvd_next = pvd->pvd_next;
235	else
236		pfs_vncache = pvd->pvd_next;
237	--pfs_vncache_entries;
238	mtx_unlock(&pfs_vncache_mutex);
239
240	FREE(pvd, M_PFSVNCACHE);
241	vp->v_data = NULL;
242	return (0);
243}
244
245/*
246 * Purge the cache of dead entries
247 *
248 * This is extremely inefficient due to the fact that vgone() not only
249 * indirectly modifies the vnode cache, but may also sleep.  We can
250 * neither hold pfs_vncache_mutex across a vgone() call, nor make any
251 * assumptions about the state of the cache after vgone() returns.  In
252 * consequence, we must start over after every vgone() call, and keep
253 * trying until we manage to traverse the entire cache.
254 *
255 * The only way to improve this situation is to change the data structure
256 * used to implement the cache.
257 */
258void
259pfs_purge(struct pfs_node *pn)
260{
261	struct pfs_vdata *pvd;
262	struct vnode *vnp;
263
264	mtx_lock(&pfs_vncache_mutex);
265	pvd = pfs_vncache;
266	while (pvd != NULL) {
267		if (pvd->pvd_dead || (pn != NULL && pvd->pvd_pn == pn)) {
268			vnp = pvd->pvd_vnode;
269			vhold(vnp);
270			mtx_unlock(&pfs_vncache_mutex);
271			VOP_LOCK(vnp, LK_EXCLUSIVE);
272			vgone(vnp);
273			VOP_UNLOCK(vnp, 0);
274			vdrop(vnp);
275			mtx_lock(&pfs_vncache_mutex);
276			pvd = pfs_vncache;
277		} else {
278			pvd = pvd->pvd_next;
279		}
280	}
281	mtx_unlock(&pfs_vncache_mutex);
282}
283
284/*
285 * Free all vnodes associated with a defunct process
286 *
287 * XXXRW: It is unfortunate that pfs_exit() always acquires and releases two
288 * mutexes (one of which is Giant) for every process exit, even if procfs
289 * isn't mounted.
290 */
291static void
292pfs_exit(void *arg, struct proc *p)
293{
294	struct pfs_vdata *pvd;
295	int dead;
296
297	if (pfs_vncache == NULL)
298		return;
299	mtx_lock(&Giant);
300	mtx_lock(&pfs_vncache_mutex);
301	for (pvd = pfs_vncache, dead = 0; pvd != NULL; pvd = pvd->pvd_next)
302		if (pvd->pvd_pid == p->p_pid)
303			dead = pvd->pvd_dead = 1;
304	mtx_unlock(&pfs_vncache_mutex);
305	if (dead)
306		pfs_purge(NULL);
307	mtx_unlock(&Giant);
308}
309