pseudofs_vncache.c revision 165737
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 165737 2007-01-02 17:27:52Z jhb $"); 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 mtx_init(&pfs_vncache_mutex, "pseudofs_vncache", NULL, MTX_DEF); 87 pfs_exit_tag = EVENTHANDLER_REGISTER(process_exit, pfs_exit, NULL, 88 EVENTHANDLER_PRI_ANY); 89} 90 91/* 92 * Tear down vnode cache 93 */ 94void 95pfs_vncache_unload(void) 96{ 97 EVENTHANDLER_DEREGISTER(process_exit, pfs_exit_tag); 98 if (pfs_vncache_entries != 0) 99 printf("pfs_vncache_unload(): %d entries remaining\n", 100 pfs_vncache_entries); 101 mtx_destroy(&pfs_vncache_mutex); 102} 103 104/* 105 * Allocate a vnode 106 */ 107int 108pfs_vncache_alloc(struct mount *mp, struct vnode **vpp, 109 struct pfs_node *pn, pid_t pid) 110{ 111 struct pfs_vdata *pvd; 112 struct vnode *vp; 113 int error; 114 115 /* 116 * See if the vnode is in the cache. 117 * XXX linear search is not very efficient. 118 */ 119retry: 120 mtx_lock(&pfs_vncache_mutex); 121 for (pvd = pfs_vncache; pvd; pvd = pvd->pvd_next) { 122 if (pvd->pvd_pn == pn && pvd->pvd_pid == pid && 123 pvd->pvd_vnode->v_mount == mp) { 124 vp = pvd->pvd_vnode; 125 VI_LOCK(vp); 126 mtx_unlock(&pfs_vncache_mutex); 127 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, curthread) == 0) { 128 ++pfs_vncache_hits; 129 *vpp = vp; 130 /* XXX see comment at top of pfs_lookup() */ 131 cache_purge(vp); 132 return (0); 133 } 134 goto retry; 135 } 136 } 137 mtx_unlock(&pfs_vncache_mutex); 138 ++pfs_vncache_misses; 139 140 /* nope, get a new one */ 141 MALLOC(pvd, struct pfs_vdata *, sizeof *pvd, M_PFSVNCACHE, M_WAITOK); 142 if (++pfs_vncache_entries > pfs_vncache_maxentries) 143 pfs_vncache_maxentries = pfs_vncache_entries; 144 error = getnewvnode("pseudofs", mp, &pfs_vnodeops, vpp); 145 if (error) { 146 FREE(pvd, M_PFSVNCACHE); 147 return (error); 148 } 149 pvd->pvd_pn = pn; 150 pvd->pvd_pid = pid; 151 (*vpp)->v_data = pvd; 152 switch (pn->pn_type) { 153 case pfstype_root: 154 (*vpp)->v_vflag = VV_ROOT; 155#if 0 156 printf("root vnode allocated\n"); 157#endif 158 /* fall through */ 159 case pfstype_dir: 160 case pfstype_this: 161 case pfstype_parent: 162 case pfstype_procdir: 163 (*vpp)->v_type = VDIR; 164 break; 165 case pfstype_file: 166 (*vpp)->v_type = VREG; 167 break; 168 case pfstype_symlink: 169 (*vpp)->v_type = VLNK; 170 break; 171 case pfstype_none: 172 KASSERT(0, ("pfs_vncache_alloc called for null node\n")); 173 default: 174 panic("%s has unexpected type: %d", pn->pn_name, pn->pn_type); 175 } 176 /* 177 * Propagate flag through to vnode so users know it can change 178 * if the process changes (i.e. execve) 179 */ 180 if ((pn->pn_flags & PFS_PROCDEP) != 0) 181 (*vpp)->v_vflag |= VV_PROCDEP; 182 pvd->pvd_vnode = *vpp; 183 mtx_lock(&pfs_vncache_mutex); 184 pvd->pvd_prev = NULL; 185 pvd->pvd_next = pfs_vncache; 186 if (pvd->pvd_next) 187 pvd->pvd_next->pvd_prev = pvd; 188 pfs_vncache = pvd; 189 mtx_unlock(&pfs_vncache_mutex); 190 (*vpp)->v_vnlock->lk_flags |= LK_CANRECURSE; 191 vn_lock(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread); 192 return (0); 193} 194 195/* 196 * Free a vnode 197 */ 198int 199pfs_vncache_free(struct vnode *vp) 200{ 201 struct pfs_vdata *pvd; 202 203 mtx_lock(&pfs_vncache_mutex); 204 pvd = (struct pfs_vdata *)vp->v_data; 205 KASSERT(pvd != NULL, ("pfs_vncache_free(): no vnode data\n")); 206 if (pvd->pvd_next) 207 pvd->pvd_next->pvd_prev = pvd->pvd_prev; 208 if (pvd->pvd_prev) 209 pvd->pvd_prev->pvd_next = pvd->pvd_next; 210 else 211 pfs_vncache = pvd->pvd_next; 212 mtx_unlock(&pfs_vncache_mutex); 213 214 --pfs_vncache_entries; 215 FREE(pvd, M_PFSVNCACHE); 216 vp->v_data = NULL; 217 return (0); 218} 219 220/* 221 * Free all vnodes associated with a defunct process 222 * 223 * XXXRW: It is unfortunate that pfs_exit() always acquires and releases two 224 * mutexes (one of which is Giant) for every process exit, even if procfs 225 * isn't mounted. 226 */ 227static void 228pfs_exit(void *arg, struct proc *p) 229{ 230 struct pfs_vdata *pvd; 231 struct vnode *vnp; 232 233 if (pfs_vncache == NULL) 234 return; 235 mtx_lock(&Giant); 236 /* 237 * This is extremely inefficient due to the fact that vgone() not 238 * only indirectly modifies the vnode cache, but may also sleep. 239 * We can neither hold pfs_vncache_mutex across a vgone() call, 240 * nor make any assumptions about the state of the cache after 241 * vgone() returns. In consequence, we must start over after 242 * every vgone() call, and keep trying until we manage to traverse 243 * the entire cache. 244 * 245 * The only way to improve this situation is to change the data 246 * structure used to implement the cache. An obvious choice in 247 * this particular case would be a BST sorted by PID. 248 */ 249 mtx_lock(&pfs_vncache_mutex); 250 pvd = pfs_vncache; 251 while (pvd != NULL) { 252 if (pvd->pvd_pid == p->p_pid) { 253 vnp = pvd->pvd_vnode; 254 vhold(vnp); 255 mtx_unlock(&pfs_vncache_mutex); 256 VOP_LOCK(vnp, LK_EXCLUSIVE, curthread); 257 vgone(vnp); 258 VOP_UNLOCK(vnp, 0, curthread); 259 vdrop(vnp); 260 mtx_lock(&pfs_vncache_mutex); 261 pvd = pfs_vncache; 262 } else { 263 pvd = pvd->pvd_next; 264 } 265 } 266 mtx_unlock(&pfs_vncache_mutex); 267 mtx_unlock(&Giant); 268} 269 270/* 271 * Disable a pseudofs node, and free all vnodes associated with it 272 */ 273int 274pfs_disable(struct pfs_node *pn) 275{ 276 struct pfs_vdata *pvd; 277 struct vnode *vnp; 278 279 if (pn->pn_flags & PFS_DISABLED) 280 return (0); 281 pn->pn_flags |= PFS_DISABLED; 282 /* XXX see comment above nearly identical code in pfs_exit() */ 283 mtx_lock(&pfs_vncache_mutex); 284 pvd = pfs_vncache; 285 while (pvd != NULL) { 286 if (pvd->pvd_pn == pn) { 287 vnp = pvd->pvd_vnode; 288 vhold(vnp); 289 mtx_unlock(&pfs_vncache_mutex); 290 VOP_LOCK(vnp, LK_EXCLUSIVE, curthread); 291 vgone(vnp); 292 VOP_UNLOCK(vnp, 0, curthread); 293 vdrop(vnp); 294 mtx_lock(&pfs_vncache_mutex); 295 pvd = pfs_vncache; 296 } else { 297 pvd = pvd->pvd_next; 298 } 299 } 300 mtx_unlock(&pfs_vncache_mutex); 301 return (0); 302} 303 304/* 305 * Re-enable a disabled pseudofs node 306 */ 307int 308pfs_enable(struct pfs_node *pn) 309{ 310 pn->pn_flags &= ~PFS_DISABLED; 311 return (0); 312} 313