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: stable/11/sys/fs/pseudofs/pseudofs_vncache.c 312248 2017-01-16 00:43:57Z kib $");
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);
54static void pfs_purge_locked(struct pfs_node *pn, bool force);
55
56static SYSCTL_NODE(_vfs_pfs, OID_AUTO, vncache, CTLFLAG_RW, 0,
57    "pseudofs vnode cache");
58
59static int pfs_vncache_entries;
60SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, entries, CTLFLAG_RD,
61    &pfs_vncache_entries, 0,
62    "number of entries in the vnode cache");
63
64static int pfs_vncache_maxentries;
65SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, maxentries, CTLFLAG_RD,
66    &pfs_vncache_maxentries, 0,
67    "highest number of entries in the vnode cache");
68
69static int pfs_vncache_hits;
70SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, hits, CTLFLAG_RD,
71    &pfs_vncache_hits, 0,
72    "number of cache hits since initialization");
73
74static int pfs_vncache_misses;
75SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, misses, CTLFLAG_RD,
76    &pfs_vncache_misses, 0,
77    "number of cache misses since initialization");
78
79extern struct vop_vector pfs_vnodeops;	/* XXX -> .h file */
80
81/*
82 * Initialize vnode cache
83 */
84void
85pfs_vncache_load(void)
86{
87
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	EVENTHANDLER_DEREGISTER(process_exit, pfs_exit_tag);
101	mtx_lock(&pfs_vncache_mutex);
102	pfs_purge_locked(NULL, true);
103	mtx_unlock(&pfs_vncache_mutex);
104	KASSERT(pfs_vncache_entries == 0,
105	    ("%d vncache entries remaining", pfs_vncache_entries));
106	mtx_destroy(&pfs_vncache_mutex);
107}
108
109/*
110 * Allocate a vnode
111 */
112int
113pfs_vncache_alloc(struct mount *mp, struct vnode **vpp,
114		  struct pfs_node *pn, pid_t pid)
115{
116	struct pfs_vdata *pvd, *pvd2;
117	struct vnode *vp;
118	int error;
119
120	/*
121	 * See if the vnode is in the cache.
122	 * XXX linear search is not very efficient.
123	 */
124retry:
125	mtx_lock(&pfs_vncache_mutex);
126	for (pvd = pfs_vncache; pvd; pvd = pvd->pvd_next) {
127		if (pvd->pvd_pn == pn && pvd->pvd_pid == pid &&
128		    pvd->pvd_vnode->v_mount == mp) {
129			vp = pvd->pvd_vnode;
130			VI_LOCK(vp);
131			mtx_unlock(&pfs_vncache_mutex);
132			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
133				++pfs_vncache_hits;
134				*vpp = vp;
135				/*
136				 * Some callers cache_enter(vp) later, so
137				 * we have to make sure it's not in the
138				 * VFS cache so it doesn't get entered
139				 * twice.  A better solution would be to
140				 * make pfs_vncache_alloc() responsible
141				 * for entering the vnode in the VFS
142				 * cache.
143				 */
144				cache_purge(vp);
145				return (0);
146			}
147			goto retry;
148		}
149	}
150	mtx_unlock(&pfs_vncache_mutex);
151
152	/* nope, get a new one */
153	pvd = malloc(sizeof *pvd, M_PFSVNCACHE, M_WAITOK);
154	pvd->pvd_next = pvd->pvd_prev = NULL;
155	error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp);
156	if (error) {
157		free(pvd, M_PFSVNCACHE);
158		return (error);
159	}
160	pvd->pvd_pn = pn;
161	pvd->pvd_pid = pid;
162	(*vpp)->v_data = pvd;
163	switch (pn->pn_type) {
164	case pfstype_root:
165		(*vpp)->v_vflag = VV_ROOT;
166#if 0
167		printf("root vnode allocated\n");
168#endif
169		/* fall through */
170	case pfstype_dir:
171	case pfstype_this:
172	case pfstype_parent:
173	case pfstype_procdir:
174		(*vpp)->v_type = VDIR;
175		break;
176	case pfstype_file:
177		(*vpp)->v_type = VREG;
178		break;
179	case pfstype_symlink:
180		(*vpp)->v_type = VLNK;
181		break;
182	case pfstype_none:
183		KASSERT(0, ("pfs_vncache_alloc called for null node\n"));
184	default:
185		panic("%s has unexpected type: %d", pn->pn_name, pn->pn_type);
186	}
187	/*
188	 * Propagate flag through to vnode so users know it can change
189	 * if the process changes (i.e. execve)
190	 */
191	if ((pn->pn_flags & PFS_PROCDEP) != 0)
192		(*vpp)->v_vflag |= VV_PROCDEP;
193	pvd->pvd_vnode = *vpp;
194	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
195	VN_LOCK_AREC(*vpp);
196	error = insmntque(*vpp, mp);
197	if (error != 0) {
198		free(pvd, M_PFSVNCACHE);
199		*vpp = NULLVP;
200		return (error);
201	}
202retry2:
203	mtx_lock(&pfs_vncache_mutex);
204	/*
205	 * Other thread may race with us, creating the entry we are
206	 * going to insert into the cache. Recheck after
207	 * pfs_vncache_mutex is reacquired.
208	 */
209	for (pvd2 = pfs_vncache; pvd2; pvd2 = pvd2->pvd_next) {
210		if (pvd2->pvd_pn == pn && pvd2->pvd_pid == pid &&
211		    pvd2->pvd_vnode->v_mount == mp) {
212			vp = pvd2->pvd_vnode;
213			VI_LOCK(vp);
214			mtx_unlock(&pfs_vncache_mutex);
215			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
216				++pfs_vncache_hits;
217				vgone(*vpp);
218				vput(*vpp);
219				*vpp = vp;
220				cache_purge(vp);
221				return (0);
222			}
223			goto retry2;
224		}
225	}
226	++pfs_vncache_misses;
227	if (++pfs_vncache_entries > pfs_vncache_maxentries)
228		pfs_vncache_maxentries = pfs_vncache_entries;
229	pvd->pvd_prev = NULL;
230	pvd->pvd_next = pfs_vncache;
231	if (pvd->pvd_next)
232		pvd->pvd_next->pvd_prev = pvd;
233	pfs_vncache = pvd;
234	mtx_unlock(&pfs_vncache_mutex);
235	return (0);
236}
237
238/*
239 * Free a vnode
240 */
241int
242pfs_vncache_free(struct vnode *vp)
243{
244	struct pfs_vdata *pvd;
245
246	mtx_lock(&pfs_vncache_mutex);
247	pvd = (struct pfs_vdata *)vp->v_data;
248	KASSERT(pvd != NULL, ("pfs_vncache_free(): no vnode data\n"));
249	if (pvd->pvd_next)
250		pvd->pvd_next->pvd_prev = pvd->pvd_prev;
251	if (pvd->pvd_prev) {
252		pvd->pvd_prev->pvd_next = pvd->pvd_next;
253		--pfs_vncache_entries;
254	} else if (pfs_vncache == pvd) {
255		pfs_vncache = pvd->pvd_next;
256		--pfs_vncache_entries;
257	}
258	mtx_unlock(&pfs_vncache_mutex);
259
260	free(pvd, M_PFSVNCACHE);
261	vp->v_data = NULL;
262	return (0);
263}
264
265/*
266 * Purge the cache of dead entries
267 *
268 * This is extremely inefficient due to the fact that vgone() not only
269 * indirectly modifies the vnode cache, but may also sleep.  We can
270 * neither hold pfs_vncache_mutex across a vgone() call, nor make any
271 * assumptions about the state of the cache after vgone() returns.  In
272 * consequence, we must start over after every vgone() call, and keep
273 * trying until we manage to traverse the entire cache.
274 *
275 * The only way to improve this situation is to change the data structure
276 * used to implement the cache.
277 */
278static void
279pfs_purge_locked(struct pfs_node *pn, bool force)
280{
281	struct pfs_vdata *pvd;
282	struct vnode *vnp;
283
284	mtx_assert(&pfs_vncache_mutex, MA_OWNED);
285	pvd = pfs_vncache;
286	while (pvd != NULL) {
287		if (force || pvd->pvd_dead ||
288		    (pn != NULL && pvd->pvd_pn == pn)) {
289			vnp = pvd->pvd_vnode;
290			vhold(vnp);
291			mtx_unlock(&pfs_vncache_mutex);
292			VOP_LOCK(vnp, LK_EXCLUSIVE);
293			vgone(vnp);
294			VOP_UNLOCK(vnp, 0);
295			mtx_lock(&pfs_vncache_mutex);
296			vdrop(vnp);
297			pvd = pfs_vncache;
298		} else {
299			pvd = pvd->pvd_next;
300		}
301	}
302}
303
304void
305pfs_purge(struct pfs_node *pn)
306{
307
308	mtx_lock(&pfs_vncache_mutex);
309	pfs_purge_locked(pn, false);
310	mtx_unlock(&pfs_vncache_mutex);
311}
312
313/*
314 * Free all vnodes associated with a defunct process
315 */
316static void
317pfs_exit(void *arg, struct proc *p)
318{
319	struct pfs_vdata *pvd;
320	int dead;
321
322	if (pfs_vncache == NULL)
323		return;
324	mtx_lock(&pfs_vncache_mutex);
325	for (pvd = pfs_vncache, dead = 0; pvd != NULL; pvd = pvd->pvd_next)
326		if (pvd->pvd_pid == p->p_pid)
327			dead = pvd->pvd_dead = 1;
328	if (dead)
329		pfs_purge_locked(NULL, false);
330	mtx_unlock(&pfs_vncache_mutex);
331}
332