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("$FreeBSD$"); 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 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_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 */ 96 void 97 pfs_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 */ 110 int 111 pfs_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 */ 122 retry: 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 pvd->pvd_next = pvd->pvd_prev = NULL; 153 error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp); 154 if (error) { 155 free(pvd, M_PFSVNCACHE); 156 return (error); 157 } 158 pvd->pvd_pn = pn; 159 pvd->pvd_pid = pid; 160 (*vpp)->v_data = pvd; 161 switch (pn->pn_type) { 162 case pfstype_root: 163 (*vpp)->v_vflag = VV_ROOT; 164 #if 0 165 printf("root vnode allocated\n"); 166 #endif 167 /* fall through */ 168 case pfstype_dir: 169 case pfstype_this: 170 case pfstype_parent: 171 case pfstype_procdir: 172 (*vpp)->v_type = VDIR; 173 break; 174 case pfstype_file: 175 (*vpp)->v_type = VREG; 176 break; 177 case pfstype_symlink: 178 (*vpp)->v_type = VLNK; 179 break; 180 case pfstype_none: 181 KASSERT(0, ("pfs_vncache_alloc called for null node\n")); 182 default: 183 panic("%s has unexpected type: %d", pn->pn_name, pn->pn_type); 184 } 185 /* 186 * Propagate flag through to vnode so users know it can change 187 * if the process changes (i.e. execve) 188 */ 189 if ((pn->pn_flags & PFS_PROCDEP) != 0) 190 (*vpp)->v_vflag |= VV_PROCDEP; 191 pvd->pvd_vnode = *vpp; 192 VN_LOCK_AREC(*vpp); 193 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY); 194 error = insmntque(*vpp, mp); 195 if (error != 0) { 196 *vpp = NULLVP; 197 return (error); 198 } 199 retry2: 200 mtx_lock(&pfs_vncache_mutex); 201 /* 202 * Other thread may race with us, creating the entry we are 203 * going to insert into the cache. Recheck after 204 * pfs_vncache_mutex is reacquired. 205 */ 206 for (pvd2 = pfs_vncache; pvd2; pvd2 = pvd2->pvd_next) { 207 if (pvd2->pvd_pn == pn && pvd2->pvd_pid == pid && 208 pvd2->pvd_vnode->v_mount == mp) { 209 vp = pvd2->pvd_vnode; 210 VI_LOCK(vp); 211 mtx_unlock(&pfs_vncache_mutex); 212 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) { 213 ++pfs_vncache_hits; 214 vgone(*vpp); 215 vput(*vpp); 216 *vpp = vp; 217 cache_purge(vp); 218 return (0); 219 } 220 goto retry2; 221 } 222 } 223 ++pfs_vncache_misses; 224 if (++pfs_vncache_entries > pfs_vncache_maxentries) 225 pfs_vncache_maxentries = pfs_vncache_entries; 226 pvd->pvd_prev = NULL; 227 pvd->pvd_next = pfs_vncache; 228 if (pvd->pvd_next) 229 pvd->pvd_next->pvd_prev = pvd; 230 pfs_vncache = pvd; 231 mtx_unlock(&pfs_vncache_mutex); 232 return (0); 233 } 234 235 /* 236 * Free a vnode 237 */ 238 int 239 pfs_vncache_free(struct vnode *vp) 240 { 241 struct pfs_vdata *pvd; 242 243 mtx_lock(&pfs_vncache_mutex); 244 pvd = (struct pfs_vdata *)vp->v_data; 245 KASSERT(pvd != NULL, ("pfs_vncache_free(): no vnode data\n")); 246 if (pvd->pvd_next) 247 pvd->pvd_next->pvd_prev = pvd->pvd_prev; 248 if (pvd->pvd_prev) 249 pvd->pvd_prev->pvd_next = pvd->pvd_next; 250 else if (pfs_vncache == pvd) 251 pfs_vncache = pvd->pvd_next; 252 --pfs_vncache_entries; 253 mtx_unlock(&pfs_vncache_mutex); 254 255 free(pvd, M_PFSVNCACHE); 256 vp->v_data = NULL; 257 return (0); 258 } 259 260 /* 261 * Purge the cache of dead entries 262 * 263 * This is extremely inefficient due to the fact that vgone() not only 264 * indirectly modifies the vnode cache, but may also sleep. We can 265 * neither hold pfs_vncache_mutex across a vgone() call, nor make any 266 * assumptions about the state of the cache after vgone() returns. In 267 * consequence, we must start over after every vgone() call, and keep 268 * trying until we manage to traverse the entire cache. 269 * 270 * The only way to improve this situation is to change the data structure 271 * used to implement the cache. 272 */ 273 void 274 pfs_purge(struct pfs_node *pn) 275 { 276 struct pfs_vdata *pvd; 277 struct vnode *vnp; 278 279 mtx_lock(&pfs_vncache_mutex); 280 pvd = pfs_vncache; 281 while (pvd != NULL) { 282 if (pvd->pvd_dead || (pn != NULL && pvd->pvd_pn == pn)) { 283 vnp = pvd->pvd_vnode; 284 vhold(vnp); 285 mtx_unlock(&pfs_vncache_mutex); 286 VOP_LOCK(vnp, LK_EXCLUSIVE); 287 vgone(vnp); 288 VOP_UNLOCK(vnp, 0); 289 vdrop(vnp); 290 mtx_lock(&pfs_vncache_mutex); 291 pvd = pfs_vncache; 292 } else { 293 pvd = pvd->pvd_next; 294 } 295 } 296 mtx_unlock(&pfs_vncache_mutex); 297 } 298 299 /* 300 * Free all vnodes associated with a defunct process 301 * 302 * XXXRW: It is unfortunate that pfs_exit() always acquires and releases two 303 * mutexes (one of which is Giant) for every process exit, even if procfs 304 * isn't mounted. 305 */ 306 static void 307 pfs_exit(void *arg, struct proc *p) 308 { 309 struct pfs_vdata *pvd; 310 int dead; 311 312 if (pfs_vncache == NULL) 313 return; 314 mtx_lock(&Giant); 315 mtx_lock(&pfs_vncache_mutex); 316 for (pvd = pfs_vncache, dead = 0; pvd != NULL; pvd = pvd->pvd_next) 317 if (pvd->pvd_pid == p->p_pid) 318 dead = pvd->pvd_dead = 1; 319 mtx_unlock(&pfs_vncache_mutex); 320 if (dead) 321 pfs_purge(NULL); 322 mtx_unlock(&Giant); 323 } 324