xref: /freebsd/sys/fs/pseudofs/pseudofs_vncache.c (revision 273c26a3c3bea87a241d6879abd4f991db180bf0)
1 /*-
2  * Copyright (c) 2001 Dag-Erling Coïdan Smørgrav
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("$FreeBSD$");
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 
48 static MALLOC_DEFINE(M_PFSVNCACHE, "pfs_vncache", "pseudofs vnode cache");
49 
50 static struct mtx pfs_vncache_mutex;
51 static struct pfs_vdata *pfs_vncache;
52 static eventhandler_tag pfs_exit_tag;
53 static void pfs_exit(void *arg, struct proc *p);
54 
55 static SYSCTL_NODE(_vfs_pfs, OID_AUTO, vncache, CTLFLAG_RW, 0,
56     "pseudofs vnode cache");
57 
58 static int pfs_vncache_entries;
59 SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, entries, CTLFLAG_RD,
60     &pfs_vncache_entries, 0,
61     "number of entries in the vnode cache");
62 
63 static int pfs_vncache_maxentries;
64 SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, maxentries, CTLFLAG_RD,
65     &pfs_vncache_maxentries, 0,
66     "highest number of entries in the vnode cache");
67 
68 static int pfs_vncache_hits;
69 SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, hits, CTLFLAG_RD,
70     &pfs_vncache_hits, 0,
71     "number of cache hits since initialization");
72 
73 static int pfs_vncache_misses;
74 SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, misses, CTLFLAG_RD,
75     &pfs_vncache_misses, 0,
76     "number of cache misses since initialization");
77 
78 extern struct vop_vector pfs_vnodeops;	/* XXX -> .h file */
79 
80 /*
81  * Initialize vnode cache
82  */
83 void
84 pfs_vncache_load(void)
85 {
86 
87 	mtx_init(&pfs_vncache_mutex, "pfs_vncache", NULL, MTX_DEF);
88 	pfs_exit_tag = EVENTHANDLER_REGISTER(process_exit, pfs_exit, NULL,
89 	    EVENTHANDLER_PRI_ANY);
90 }
91 
92 /*
93  * Tear down vnode cache
94  */
95 void
96 pfs_vncache_unload(void)
97 {
98 
99 	EVENTHANDLER_DEREGISTER(process_exit, pfs_exit_tag);
100 	KASSERT(pfs_vncache_entries == 0,
101 	    ("%d vncache entries remaining", pfs_vncache_entries));
102 	mtx_destroy(&pfs_vncache_mutex);
103 }
104 
105 /*
106  * Allocate a vnode
107  */
108 int
109 pfs_vncache_alloc(struct mount *mp, struct vnode **vpp,
110 		  struct pfs_node *pn, pid_t pid)
111 {
112 	struct pfs_vdata *pvd, *pvd2;
113 	struct vnode *vp;
114 	int error;
115 
116 	/*
117 	 * See if the vnode is in the cache.
118 	 * XXX linear search is not very efficient.
119 	 */
120 retry:
121 	mtx_lock(&pfs_vncache_mutex);
122 	for (pvd = pfs_vncache; pvd; pvd = pvd->pvd_next) {
123 		if (pvd->pvd_pn == pn && pvd->pvd_pid == pid &&
124 		    pvd->pvd_vnode->v_mount == mp) {
125 			vp = pvd->pvd_vnode;
126 			VI_LOCK(vp);
127 			mtx_unlock(&pfs_vncache_mutex);
128 			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
129 				++pfs_vncache_hits;
130 				*vpp = vp;
131 				/*
132 				 * Some callers cache_enter(vp) later, so
133 				 * we have to make sure it's not in the
134 				 * VFS cache so it doesn't get entered
135 				 * twice.  A better solution would be to
136 				 * make pfs_vncache_alloc() responsible
137 				 * for entering the vnode in the VFS
138 				 * cache.
139 				 */
140 				cache_purge(vp);
141 				return (0);
142 			}
143 			goto retry;
144 		}
145 	}
146 	mtx_unlock(&pfs_vncache_mutex);
147 
148 	/* nope, get a new one */
149 	pvd = malloc(sizeof *pvd, M_PFSVNCACHE, M_WAITOK);
150 	pvd->pvd_next = pvd->pvd_prev = NULL;
151 	error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp);
152 	if (error) {
153 		free(pvd, M_PFSVNCACHE);
154 		return (error);
155 	}
156 	pvd->pvd_pn = pn;
157 	pvd->pvd_pid = pid;
158 	(*vpp)->v_data = pvd;
159 	switch (pn->pn_type) {
160 	case pfstype_root:
161 		(*vpp)->v_vflag = VV_ROOT;
162 #if 0
163 		printf("root vnode allocated\n");
164 #endif
165 		/* fall through */
166 	case pfstype_dir:
167 	case pfstype_this:
168 	case pfstype_parent:
169 	case pfstype_procdir:
170 		(*vpp)->v_type = VDIR;
171 		break;
172 	case pfstype_file:
173 		(*vpp)->v_type = VREG;
174 		break;
175 	case pfstype_symlink:
176 		(*vpp)->v_type = VLNK;
177 		break;
178 	case pfstype_none:
179 		KASSERT(0, ("pfs_vncache_alloc called for null node\n"));
180 	default:
181 		panic("%s has unexpected type: %d", pn->pn_name, pn->pn_type);
182 	}
183 	/*
184 	 * Propagate flag through to vnode so users know it can change
185 	 * if the process changes (i.e. execve)
186 	 */
187 	if ((pn->pn_flags & PFS_PROCDEP) != 0)
188 		(*vpp)->v_vflag |= VV_PROCDEP;
189 	pvd->pvd_vnode = *vpp;
190 	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
191 	VN_LOCK_AREC(*vpp);
192 	error = insmntque(*vpp, mp);
193 	if (error != 0) {
194 		free(pvd, M_PFSVNCACHE);
195 		*vpp = NULLVP;
196 		return (error);
197 	}
198 retry2:
199 	mtx_lock(&pfs_vncache_mutex);
200 	/*
201 	 * Other thread may race with us, creating the entry we are
202 	 * going to insert into the cache. Recheck after
203 	 * pfs_vncache_mutex is reacquired.
204 	 */
205 	for (pvd2 = pfs_vncache; pvd2; pvd2 = pvd2->pvd_next) {
206 		if (pvd2->pvd_pn == pn && pvd2->pvd_pid == pid &&
207 		    pvd2->pvd_vnode->v_mount == mp) {
208 			vp = pvd2->pvd_vnode;
209 			VI_LOCK(vp);
210 			mtx_unlock(&pfs_vncache_mutex);
211 			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
212 				++pfs_vncache_hits;
213 				vgone(*vpp);
214 				vput(*vpp);
215 				*vpp = vp;
216 				cache_purge(vp);
217 				return (0);
218 			}
219 			goto retry2;
220 		}
221 	}
222 	++pfs_vncache_misses;
223 	if (++pfs_vncache_entries > pfs_vncache_maxentries)
224 		pfs_vncache_maxentries = pfs_vncache_entries;
225 	pvd->pvd_prev = NULL;
226 	pvd->pvd_next = pfs_vncache;
227 	if (pvd->pvd_next)
228 		pvd->pvd_next->pvd_prev = pvd;
229 	pfs_vncache = pvd;
230 	mtx_unlock(&pfs_vncache_mutex);
231 	return (0);
232 }
233 
234 /*
235  * Free a vnode
236  */
237 int
238 pfs_vncache_free(struct vnode *vp)
239 {
240 	struct pfs_vdata *pvd;
241 
242 	mtx_lock(&pfs_vncache_mutex);
243 	pvd = (struct pfs_vdata *)vp->v_data;
244 	KASSERT(pvd != NULL, ("pfs_vncache_free(): no vnode data\n"));
245 	if (pvd->pvd_next)
246 		pvd->pvd_next->pvd_prev = pvd->pvd_prev;
247 	if (pvd->pvd_prev) {
248 		pvd->pvd_prev->pvd_next = pvd->pvd_next;
249 		--pfs_vncache_entries;
250 	} else if (pfs_vncache == pvd) {
251 		pfs_vncache = pvd->pvd_next;
252 		--pfs_vncache_entries;
253 	}
254 	mtx_unlock(&pfs_vncache_mutex);
255 
256 	free(pvd, M_PFSVNCACHE);
257 	vp->v_data = NULL;
258 	return (0);
259 }
260 
261 /*
262  * Purge the cache of dead entries
263  *
264  * This is extremely inefficient due to the fact that vgone() not only
265  * indirectly modifies the vnode cache, but may also sleep.  We can
266  * neither hold pfs_vncache_mutex across a vgone() call, nor make any
267  * assumptions about the state of the cache after vgone() returns.  In
268  * consequence, we must start over after every vgone() call, and keep
269  * trying until we manage to traverse the entire cache.
270  *
271  * The only way to improve this situation is to change the data structure
272  * used to implement the cache.
273  */
274 static void
275 pfs_purge_locked(struct pfs_node *pn)
276 {
277 	struct pfs_vdata *pvd;
278 	struct vnode *vnp;
279 
280 	mtx_assert(&pfs_vncache_mutex, MA_OWNED);
281 	pvd = pfs_vncache;
282 	while (pvd != NULL) {
283 		if (pvd->pvd_dead || (pn != NULL && pvd->pvd_pn == pn)) {
284 			vnp = pvd->pvd_vnode;
285 			vhold(vnp);
286 			mtx_unlock(&pfs_vncache_mutex);
287 			VOP_LOCK(vnp, LK_EXCLUSIVE);
288 			vgone(vnp);
289 			VOP_UNLOCK(vnp, 0);
290 			mtx_lock(&pfs_vncache_mutex);
291 			vdrop(vnp);
292 			pvd = pfs_vncache;
293 		} else {
294 			pvd = pvd->pvd_next;
295 		}
296 	}
297 }
298 
299 void
300 pfs_purge(struct pfs_node *pn)
301 {
302 
303 	mtx_lock(&pfs_vncache_mutex);
304 	pfs_purge_locked(pn);
305 	mtx_unlock(&pfs_vncache_mutex);
306 }
307 
308 /*
309  * Free all vnodes associated with a defunct process
310  */
311 static void
312 pfs_exit(void *arg, struct proc *p)
313 {
314 	struct pfs_vdata *pvd;
315 	int dead;
316 
317 	if (pfs_vncache == NULL)
318 		return;
319 	mtx_lock(&pfs_vncache_mutex);
320 	for (pvd = pfs_vncache, dead = 0; pvd != NULL; pvd = pvd->pvd_next)
321 		if (pvd->pvd_pid == p->p_pid)
322 			dead = pvd->pvd_dead = 1;
323 	if (dead)
324 		pfs_purge_locked(NULL);
325 	mtx_unlock(&pfs_vncache_mutex);
326 }
327