pseudofs_vncache.c revision 186560
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 186560 2008-12-29 12:07:18Z 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); 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, *pvd2; 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); 149 150 /* nope, get a new one */ 151 pvd = malloc(sizeof *pvd, M_PFSVNCACHE, M_WAITOK); 152 error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp); 153 if (error) { 154 free(pvd, M_PFSVNCACHE); 155 return (error); 156 } 157 pvd->pvd_pn = pn; 158 pvd->pvd_pid = pid; 159 (*vpp)->v_data = pvd; 160 switch (pn->pn_type) { 161 case pfstype_root: 162 (*vpp)->v_vflag = VV_ROOT; 163#if 0 164 printf("root vnode allocated\n"); 165#endif 166 /* fall through */ 167 case pfstype_dir: 168 case pfstype_this: 169 case pfstype_parent: 170 case pfstype_procdir: 171 (*vpp)->v_type = VDIR; 172 break; 173 case pfstype_file: 174 (*vpp)->v_type = VREG; 175 break; 176 case pfstype_symlink: 177 (*vpp)->v_type = VLNK; 178 break; 179 case pfstype_none: 180 KASSERT(0, ("pfs_vncache_alloc called for null node\n")); 181 default: 182 panic("%s has unexpected type: %d", pn->pn_name, pn->pn_type); 183 } 184 /* 185 * Propagate flag through to vnode so users know it can change 186 * if the process changes (i.e. execve) 187 */ 188 if ((pn->pn_flags & PFS_PROCDEP) != 0) 189 (*vpp)->v_vflag |= VV_PROCDEP; 190 pvd->pvd_vnode = *vpp; 191 VN_LOCK_AREC(*vpp); 192 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 193 error = insmntque(*vpp, mp); 194 if (error != 0) { 195 *vpp = NULLVP; 196 return (error); 197 } 198retry2: 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 *vpp = vp; 215 cache_purge(vp); 216 return (0); 217 } 218 goto retry2; 219 } 220 } 221 ++pfs_vncache_misses; 222 if (++pfs_vncache_entries > pfs_vncache_maxentries) 223 pfs_vncache_maxentries = pfs_vncache_entries; 224 pvd->pvd_prev = NULL; 225 pvd->pvd_next = pfs_vncache; 226 if (pvd->pvd_next) 227 pvd->pvd_next->pvd_prev = pvd; 228 pfs_vncache = pvd; 229 mtx_unlock(&pfs_vncache_mutex); 230 return (0); 231} 232 233/* 234 * Free a vnode 235 */ 236int 237pfs_vncache_free(struct vnode *vp) 238{ 239 struct pfs_vdata *pvd; 240 241 mtx_lock(&pfs_vncache_mutex); 242 pvd = (struct pfs_vdata *)vp->v_data; 243 KASSERT(pvd != NULL, ("pfs_vncache_free(): no vnode data\n")); 244 if (pvd->pvd_next) 245 pvd->pvd_next->pvd_prev = pvd->pvd_prev; 246 if (pvd->pvd_prev) 247 pvd->pvd_prev->pvd_next = pvd->pvd_next; 248 else 249 pfs_vncache = pvd->pvd_next; 250 --pfs_vncache_entries; 251 mtx_unlock(&pfs_vncache_mutex); 252 253 free(pvd, M_PFSVNCACHE); 254 vp->v_data = NULL; 255 return (0); 256} 257 258/* 259 * Purge the cache of dead entries 260 * 261 * This is extremely inefficient due to the fact that vgone() not only 262 * indirectly modifies the vnode cache, but may also sleep. We can 263 * neither hold pfs_vncache_mutex across a vgone() call, nor make any 264 * assumptions about the state of the cache after vgone() returns. In 265 * consequence, we must start over after every vgone() call, and keep 266 * trying until we manage to traverse the entire cache. 267 * 268 * The only way to improve this situation is to change the data structure 269 * used to implement the cache. 270 */ 271void 272pfs_purge(struct pfs_node *pn) 273{ 274 struct pfs_vdata *pvd; 275 struct vnode *vnp; 276 277 mtx_lock(&pfs_vncache_mutex); 278 pvd = pfs_vncache; 279 while (pvd != NULL) { 280 if (pvd->pvd_dead || (pn != NULL && pvd->pvd_pn == pn)) { 281 vnp = pvd->pvd_vnode; 282 vhold(vnp); 283 mtx_unlock(&pfs_vncache_mutex); 284 VOP_LOCK(vnp, LK_EXCLUSIVE); 285 vgone(vnp); 286 VOP_UNLOCK(vnp, 0); 287 vdrop(vnp); 288 mtx_lock(&pfs_vncache_mutex); 289 pvd = pfs_vncache; 290 } else { 291 pvd = pvd->pvd_next; 292 } 293 } 294 mtx_unlock(&pfs_vncache_mutex); 295} 296 297/* 298 * Free all vnodes associated with a defunct process 299 * 300 * XXXRW: It is unfortunate that pfs_exit() always acquires and releases two 301 * mutexes (one of which is Giant) for every process exit, even if procfs 302 * isn't mounted. 303 */ 304static void 305pfs_exit(void *arg, struct proc *p) 306{ 307 struct pfs_vdata *pvd; 308 int dead; 309 310 if (pfs_vncache == NULL) 311 return; 312 mtx_lock(&Giant); 313 mtx_lock(&pfs_vncache_mutex); 314 for (pvd = pfs_vncache, dead = 0; pvd != NULL; pvd = pvd->pvd_next) 315 if (pvd->pvd_pid == p->p_pid) 316 dead = pvd->pvd_dead = 1; 317 mtx_unlock(&pfs_vncache_mutex); 318 if (dead) 319 pfs_purge(NULL); 320 mtx_unlock(&Giant); 321} 322