xref: /freebsd/sys/fs/pseudofs/pseudofs_vncache.c (revision 68d75eff68281c1b445e3010bb975eae07aac225)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2001 Dag-Erling Coïdan Smørgrav
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer
12  *    in this position and unchanged.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. The name of the author may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_pseudofs.h"
35 
36 #include <sys/param.h>
37 #include <sys/kernel.h>
38 #include <sys/systm.h>
39 #include <sys/eventhandler.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mutex.h>
43 #include <sys/proc.h>
44 #include <sys/sysctl.h>
45 #include <sys/vnode.h>
46 
47 #include <fs/pseudofs/pseudofs.h>
48 #include <fs/pseudofs/pseudofs_internal.h>
49 
50 static MALLOC_DEFINE(M_PFSVNCACHE, "pfs_vncache", "pseudofs vnode cache");
51 
52 static struct mtx pfs_vncache_mutex;
53 static eventhandler_tag pfs_exit_tag;
54 static void pfs_exit(void *arg, struct proc *p);
55 static void pfs_purge_all(void);
56 
57 static SYSCTL_NODE(_vfs_pfs, OID_AUTO, vncache, CTLFLAG_RW, 0,
58     "pseudofs vnode cache");
59 
60 static int pfs_vncache_entries;
61 SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, entries, CTLFLAG_RD,
62     &pfs_vncache_entries, 0,
63     "number of entries in the vnode cache");
64 
65 static int pfs_vncache_maxentries;
66 SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, maxentries, CTLFLAG_RD,
67     &pfs_vncache_maxentries, 0,
68     "highest number of entries in the vnode cache");
69 
70 static int pfs_vncache_hits;
71 SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, hits, CTLFLAG_RD,
72     &pfs_vncache_hits, 0,
73     "number of cache hits since initialization");
74 
75 static int pfs_vncache_misses;
76 SYSCTL_INT(_vfs_pfs_vncache, OID_AUTO, misses, CTLFLAG_RD,
77     &pfs_vncache_misses, 0,
78     "number of cache misses since initialization");
79 
80 extern struct vop_vector pfs_vnodeops;	/* XXX -> .h file */
81 
82 static SLIST_HEAD(pfs_vncache_head, pfs_vdata) *pfs_vncache_hashtbl;
83 static u_long pfs_vncache_hash;
84 #define PFS_VNCACHE_HASH(pid)	(&pfs_vncache_hashtbl[(pid) & pfs_vncache_hash])
85 
86 /*
87  * Initialize vnode cache
88  */
89 void
90 pfs_vncache_load(void)
91 {
92 
93 	mtx_init(&pfs_vncache_mutex, "pfs_vncache", NULL, MTX_DEF);
94 	pfs_vncache_hashtbl = hashinit(maxproc / 4, M_PFSVNCACHE, &pfs_vncache_hash);
95 	pfs_exit_tag = EVENTHANDLER_REGISTER(process_exit, pfs_exit, NULL,
96 	    EVENTHANDLER_PRI_ANY);
97 }
98 
99 /*
100  * Tear down vnode cache
101  */
102 void
103 pfs_vncache_unload(void)
104 {
105 
106 	EVENTHANDLER_DEREGISTER(process_exit, pfs_exit_tag);
107 	pfs_purge_all();
108 	KASSERT(pfs_vncache_entries == 0,
109 	    ("%d vncache entries remaining", pfs_vncache_entries));
110 	mtx_destroy(&pfs_vncache_mutex);
111 }
112 
113 /*
114  * Allocate a vnode
115  */
116 int
117 pfs_vncache_alloc(struct mount *mp, struct vnode **vpp,
118 		  struct pfs_node *pn, pid_t pid)
119 {
120 	struct pfs_vncache_head *hash;
121 	struct pfs_vdata *pvd, *pvd2;
122 	struct vnode *vp;
123 	int error;
124 
125 	/*
126 	 * See if the vnode is in the cache.
127 	 */
128 	hash = PFS_VNCACHE_HASH(pid);
129 	if (SLIST_EMPTY(hash))
130 		goto alloc;
131 retry:
132 	mtx_lock(&pfs_vncache_mutex);
133 	SLIST_FOREACH(pvd, hash, pvd_hash) {
134 		if (pvd->pvd_pn == pn && pvd->pvd_pid == pid &&
135 		    pvd->pvd_vnode->v_mount == mp) {
136 			vp = pvd->pvd_vnode;
137 			VI_LOCK(vp);
138 			mtx_unlock(&pfs_vncache_mutex);
139 			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
140 				++pfs_vncache_hits;
141 				*vpp = vp;
142 				/*
143 				 * Some callers cache_enter(vp) later, so
144 				 * we have to make sure it's not in the
145 				 * VFS cache so it doesn't get entered
146 				 * twice.  A better solution would be to
147 				 * make pfs_vncache_alloc() responsible
148 				 * for entering the vnode in the VFS
149 				 * cache.
150 				 */
151 				cache_purge(vp);
152 				return (0);
153 			}
154 			goto retry;
155 		}
156 	}
157 	mtx_unlock(&pfs_vncache_mutex);
158 alloc:
159 	/* nope, get a new one */
160 	pvd = malloc(sizeof *pvd, M_PFSVNCACHE, M_WAITOK);
161 	error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp);
162 	if (error) {
163 		free(pvd, M_PFSVNCACHE);
164 		return (error);
165 	}
166 	pvd->pvd_pn = pn;
167 	pvd->pvd_pid = pid;
168 	(*vpp)->v_data = pvd;
169 	switch (pn->pn_type) {
170 	case pfstype_root:
171 		(*vpp)->v_vflag = VV_ROOT;
172 #if 0
173 		printf("root vnode allocated\n");
174 #endif
175 		/* fall through */
176 	case pfstype_dir:
177 	case pfstype_this:
178 	case pfstype_parent:
179 	case pfstype_procdir:
180 		(*vpp)->v_type = VDIR;
181 		break;
182 	case pfstype_file:
183 		(*vpp)->v_type = VREG;
184 		break;
185 	case pfstype_symlink:
186 		(*vpp)->v_type = VLNK;
187 		break;
188 	case pfstype_none:
189 		KASSERT(0, ("pfs_vncache_alloc called for null node\n"));
190 	default:
191 		panic("%s has unexpected type: %d", pn->pn_name, pn->pn_type);
192 	}
193 	/*
194 	 * Propagate flag through to vnode so users know it can change
195 	 * if the process changes (i.e. execve)
196 	 */
197 	if ((pn->pn_flags & PFS_PROCDEP) != 0)
198 		(*vpp)->v_vflag |= VV_PROCDEP;
199 	pvd->pvd_vnode = *vpp;
200 	vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
201 	VN_LOCK_AREC(*vpp);
202 	error = insmntque(*vpp, mp);
203 	if (error != 0) {
204 		free(pvd, M_PFSVNCACHE);
205 		*vpp = NULLVP;
206 		return (error);
207 	}
208 retry2:
209 	mtx_lock(&pfs_vncache_mutex);
210 	/*
211 	 * Other thread may race with us, creating the entry we are
212 	 * going to insert into the cache. Recheck after
213 	 * pfs_vncache_mutex is reacquired.
214 	 */
215 	SLIST_FOREACH(pvd2, hash, pvd_hash) {
216 		if (pvd2->pvd_pn == pn && pvd2->pvd_pid == pid &&
217 		    pvd2->pvd_vnode->v_mount == mp) {
218 			vp = pvd2->pvd_vnode;
219 			VI_LOCK(vp);
220 			mtx_unlock(&pfs_vncache_mutex);
221 			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) {
222 				++pfs_vncache_hits;
223 				vgone(*vpp);
224 				vput(*vpp);
225 				*vpp = vp;
226 				cache_purge(vp);
227 				return (0);
228 			}
229 			goto retry2;
230 		}
231 	}
232 	++pfs_vncache_misses;
233 	if (++pfs_vncache_entries > pfs_vncache_maxentries)
234 		pfs_vncache_maxentries = pfs_vncache_entries;
235 	SLIST_INSERT_HEAD(hash, pvd, pvd_hash);
236 	mtx_unlock(&pfs_vncache_mutex);
237 	return (0);
238 }
239 
240 /*
241  * Free a vnode
242  */
243 int
244 pfs_vncache_free(struct vnode *vp)
245 {
246 	struct pfs_vdata *pvd, *pvd2;
247 
248 	mtx_lock(&pfs_vncache_mutex);
249 	pvd = (struct pfs_vdata *)vp->v_data;
250 	KASSERT(pvd != NULL, ("pfs_vncache_free(): no vnode data\n"));
251 	SLIST_FOREACH(pvd2, PFS_VNCACHE_HASH(pvd->pvd_pid), pvd_hash) {
252 		if (pvd2 != pvd)
253 			continue;
254 		SLIST_REMOVE(PFS_VNCACHE_HASH(pvd->pvd_pid), pvd, pfs_vdata, pvd_hash);
255 		--pfs_vncache_entries;
256 		break;
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  * The code is not very efficient and this perhaps can be addressed without
269  * a complete rewrite. Previous iteration was walking a linked list from
270  * scratch every time. This code only walks the relevant hash chain (if pid
271  * is provided), but still resorts to scanning the entire cache at least twice
272  * if a specific component is to be removed which is slower. This can be
273  * augmented with resizing the hash.
274  *
275  * Explanation of the previous state:
276  *
277  * This is extremely inefficient due to the fact that vgone() not only
278  * indirectly modifies the vnode cache, but may also sleep.  We can
279  * neither hold pfs_vncache_mutex across a vgone() call, nor make any
280  * assumptions about the state of the cache after vgone() returns.  In
281  * consequence, we must start over after every vgone() call, and keep
282  * trying until we manage to traverse the entire cache.
283  *
284  * The only way to improve this situation is to change the data structure
285  * used to implement the cache.
286  */
287 
288 static void
289 pfs_purge_one(struct vnode *vnp)
290 {
291 
292 	VOP_LOCK(vnp, LK_EXCLUSIVE);
293 	vgone(vnp);
294 	VOP_UNLOCK(vnp);
295 	vdrop(vnp);
296 }
297 
298 void
299 pfs_purge(struct pfs_node *pn)
300 {
301 	struct pfs_vdata *pvd;
302 	struct vnode *vnp;
303 	u_long i, removed;
304 
305 	mtx_lock(&pfs_vncache_mutex);
306 restart:
307 	removed = 0;
308 	for (i = 0; i < pfs_vncache_hash; i++) {
309 restart_chain:
310 		SLIST_FOREACH(pvd, &pfs_vncache_hashtbl[i], pvd_hash) {
311 			if (pn != NULL && pvd->pvd_pn != pn)
312 				continue;
313 			vnp = pvd->pvd_vnode;
314 			vhold(vnp);
315 			mtx_unlock(&pfs_vncache_mutex);
316 			pfs_purge_one(vnp);
317 			removed++;
318 			mtx_lock(&pfs_vncache_mutex);
319 			goto restart_chain;
320 		}
321 	}
322 	if (removed > 0)
323 		goto restart;
324 	mtx_unlock(&pfs_vncache_mutex);
325 }
326 
327 static void
328 pfs_purge_all(void)
329 {
330 
331 	pfs_purge(NULL);
332 }
333 
334 /*
335  * Free all vnodes associated with a defunct process
336  */
337 static void
338 pfs_exit(void *arg, struct proc *p)
339 {
340 	struct pfs_vncache_head *hash;
341 	struct pfs_vdata *pvd;
342 	struct vnode *vnp;
343 	int pid;
344 
345 	pid = p->p_pid;
346 	hash = PFS_VNCACHE_HASH(pid);
347 	if (SLIST_EMPTY(hash))
348 		return;
349 restart:
350 	mtx_lock(&pfs_vncache_mutex);
351 	SLIST_FOREACH(pvd, hash, pvd_hash) {
352 		if (pvd->pvd_pid != pid)
353 			continue;
354 		vnp = pvd->pvd_vnode;
355 		vhold(vnp);
356 		mtx_unlock(&pfs_vncache_mutex);
357 		pfs_purge_one(vnp);
358 		goto restart;
359 	}
360 	mtx_unlock(&pfs_vncache_mutex);
361 }
362