vfs_vnode.c revision 1.5.2.2
1 1.5.2.2 rmind /* $NetBSD: vfs_vnode.c,v 1.5.2.2 2011/04/21 01:42:12 rmind Exp $ */ 2 1.5.2.2 rmind 3 1.5.2.2 rmind /*- 4 1.5.2.2 rmind * Copyright (c) 1997-2011 The NetBSD Foundation, Inc. 5 1.5.2.2 rmind * All rights reserved. 6 1.5.2.2 rmind * 7 1.5.2.2 rmind * This code is derived from software contributed to The NetBSD Foundation 8 1.5.2.2 rmind * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 1.5.2.2 rmind * NASA Ames Research Center, by Charles M. Hannum, and by Andrew Doran. 10 1.5.2.2 rmind * 11 1.5.2.2 rmind * Redistribution and use in source and binary forms, with or without 12 1.5.2.2 rmind * modification, are permitted provided that the following conditions 13 1.5.2.2 rmind * are met: 14 1.5.2.2 rmind * 1. Redistributions of source code must retain the above copyright 15 1.5.2.2 rmind * notice, this list of conditions and the following disclaimer. 16 1.5.2.2 rmind * 2. Redistributions in binary form must reproduce the above copyright 17 1.5.2.2 rmind * notice, this list of conditions and the following disclaimer in the 18 1.5.2.2 rmind * documentation and/or other materials provided with the distribution. 19 1.5.2.2 rmind * 20 1.5.2.2 rmind * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 1.5.2.2 rmind * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 1.5.2.2 rmind * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 1.5.2.2 rmind * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 1.5.2.2 rmind * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 1.5.2.2 rmind * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 1.5.2.2 rmind * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 1.5.2.2 rmind * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 1.5.2.2 rmind * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 1.5.2.2 rmind * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 1.5.2.2 rmind * POSSIBILITY OF SUCH DAMAGE. 31 1.5.2.2 rmind */ 32 1.5.2.2 rmind 33 1.5.2.2 rmind /* 34 1.5.2.2 rmind * Copyright (c) 1989, 1993 35 1.5.2.2 rmind * The Regents of the University of California. All rights reserved. 36 1.5.2.2 rmind * (c) UNIX System Laboratories, Inc. 37 1.5.2.2 rmind * All or some portions of this file are derived from material licensed 38 1.5.2.2 rmind * to the University of California by American Telephone and Telegraph 39 1.5.2.2 rmind * Co. or Unix System Laboratories, Inc. and are reproduced herein with 40 1.5.2.2 rmind * the permission of UNIX System Laboratories, Inc. 41 1.5.2.2 rmind * 42 1.5.2.2 rmind * Redistribution and use in source and binary forms, with or without 43 1.5.2.2 rmind * modification, are permitted provided that the following conditions 44 1.5.2.2 rmind * are met: 45 1.5.2.2 rmind * 1. Redistributions of source code must retain the above copyright 46 1.5.2.2 rmind * notice, this list of conditions and the following disclaimer. 47 1.5.2.2 rmind * 2. Redistributions in binary form must reproduce the above copyright 48 1.5.2.2 rmind * notice, this list of conditions and the following disclaimer in the 49 1.5.2.2 rmind * documentation and/or other materials provided with the distribution. 50 1.5.2.2 rmind * 3. Neither the name of the University nor the names of its contributors 51 1.5.2.2 rmind * may be used to endorse or promote products derived from this software 52 1.5.2.2 rmind * without specific prior written permission. 53 1.5.2.2 rmind * 54 1.5.2.2 rmind * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 1.5.2.2 rmind * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 1.5.2.2 rmind * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 1.5.2.2 rmind * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 1.5.2.2 rmind * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 1.5.2.2 rmind * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 1.5.2.2 rmind * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 1.5.2.2 rmind * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 1.5.2.2 rmind * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 1.5.2.2 rmind * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 1.5.2.2 rmind * SUCH DAMAGE. 65 1.5.2.2 rmind * 66 1.5.2.2 rmind * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 67 1.5.2.2 rmind */ 68 1.5.2.2 rmind 69 1.5.2.2 rmind /* 70 1.5.2.2 rmind * Note on v_usecount and locking: 71 1.5.2.2 rmind * 72 1.5.2.2 rmind * At nearly all points it is known that v_usecount could be zero, the 73 1.5.2.2 rmind * vnode interlock will be held. 74 1.5.2.2 rmind * 75 1.5.2.2 rmind * To change v_usecount away from zero, the interlock must be held. To 76 1.5.2.2 rmind * change from a non-zero value to zero, again the interlock must be 77 1.5.2.2 rmind * held. 78 1.5.2.2 rmind * 79 1.5.2.2 rmind * There's a flag bit, VC_XLOCK, embedded in v_usecount. 80 1.5.2.2 rmind * To raise v_usecount, if the VC_XLOCK bit is set in it, the interlock 81 1.5.2.2 rmind * must be held. 82 1.5.2.2 rmind * To modify the VC_XLOCK bit, the interlock must be held. 83 1.5.2.2 rmind * We always keep the usecount (v_usecount & VC_MASK) non-zero while the 84 1.5.2.2 rmind * VC_XLOCK bit is set. 85 1.5.2.2 rmind * 86 1.5.2.2 rmind * Unless the VC_XLOCK bit is set, changing the usecount from a non-zero 87 1.5.2.2 rmind * value to a non-zero value can safely be done using atomic operations, 88 1.5.2.2 rmind * without the interlock held. 89 1.5.2.2 rmind * Even if the VC_XLOCK bit is set, decreasing the usecount to a non-zero 90 1.5.2.2 rmind * value can be done using atomic operations, without the interlock held. 91 1.5.2.2 rmind */ 92 1.5.2.2 rmind 93 1.5.2.2 rmind #include <sys/cdefs.h> 94 1.5.2.2 rmind __KERNEL_RCSID(0, "$NetBSD: vfs_vnode.c,v 1.5.2.2 2011/04/21 01:42:12 rmind Exp $"); 95 1.5.2.2 rmind 96 1.5.2.2 rmind #include <sys/param.h> 97 1.5.2.2 rmind #include <sys/kernel.h> 98 1.5.2.2 rmind 99 1.5.2.2 rmind #include <sys/atomic.h> 100 1.5.2.2 rmind #include <sys/buf.h> 101 1.5.2.2 rmind #include <sys/conf.h> 102 1.5.2.2 rmind #include <sys/device.h> 103 1.5.2.2 rmind #include <sys/kauth.h> 104 1.5.2.2 rmind #include <sys/kmem.h> 105 1.5.2.2 rmind #include <sys/kthread.h> 106 1.5.2.2 rmind #include <sys/module.h> 107 1.5.2.2 rmind #include <sys/mount.h> 108 1.5.2.2 rmind #include <sys/namei.h> 109 1.5.2.2 rmind #include <sys/syscallargs.h> 110 1.5.2.2 rmind #include <sys/sysctl.h> 111 1.5.2.2 rmind #include <sys/systm.h> 112 1.5.2.2 rmind #include <sys/vnode.h> 113 1.5.2.2 rmind #include <sys/wapbl.h> 114 1.5.2.2 rmind 115 1.5.2.2 rmind #include <uvm/uvm.h> 116 1.5.2.2 rmind #include <uvm/uvm_readahead.h> 117 1.5.2.2 rmind 118 1.5.2.2 rmind u_int numvnodes; 119 1.5.2.2 rmind 120 1.5.2.2 rmind static pool_cache_t vnode_cache; 121 1.5.2.2 rmind static kmutex_t vnode_free_list_lock; 122 1.5.2.2 rmind 123 1.5.2.2 rmind static vnodelst_t vnode_free_list; 124 1.5.2.2 rmind static vnodelst_t vnode_hold_list; 125 1.5.2.2 rmind static vnodelst_t vrele_list; 126 1.5.2.2 rmind 127 1.5.2.2 rmind static kmutex_t vrele_lock; 128 1.5.2.2 rmind static kcondvar_t vrele_cv; 129 1.5.2.2 rmind static lwp_t * vrele_lwp; 130 1.5.2.2 rmind static int vrele_pending; 131 1.5.2.2 rmind static int vrele_gen; 132 1.5.2.2 rmind 133 1.5.2.2 rmind static vnode_t * getcleanvnode(void); 134 1.5.2.2 rmind static void vrele_thread(void *); 135 1.5.2.2 rmind static void vpanic(vnode_t *, const char *); 136 1.5.2.2 rmind 137 1.5.2.2 rmind /* Routines having to do with the management of the vnode table. */ 138 1.5.2.2 rmind extern int (**dead_vnodeop_p)(void *); 139 1.5.2.2 rmind 140 1.5.2.2 rmind void 141 1.5.2.2 rmind vfs_vnode_sysinit(void) 142 1.5.2.2 rmind { 143 1.5.2.2 rmind int error; 144 1.5.2.2 rmind 145 1.5.2.2 rmind vnode_cache = pool_cache_init(sizeof(vnode_t), 0, 0, 0, "vnodepl", 146 1.5.2.2 rmind NULL, IPL_NONE, NULL, NULL, NULL); 147 1.5.2.2 rmind KASSERT(vnode_cache != NULL); 148 1.5.2.2 rmind 149 1.5.2.2 rmind mutex_init(&vnode_free_list_lock, MUTEX_DEFAULT, IPL_NONE); 150 1.5.2.2 rmind TAILQ_INIT(&vnode_free_list); 151 1.5.2.2 rmind TAILQ_INIT(&vnode_hold_list); 152 1.5.2.2 rmind TAILQ_INIT(&vrele_list); 153 1.5.2.2 rmind 154 1.5.2.2 rmind mutex_init(&vrele_lock, MUTEX_DEFAULT, IPL_NONE); 155 1.5.2.2 rmind cv_init(&vrele_cv, "vrele"); 156 1.5.2.2 rmind error = kthread_create(PRI_VM, KTHREAD_MPSAFE, NULL, vrele_thread, 157 1.5.2.2 rmind NULL, &vrele_lwp, "vrele"); 158 1.5.2.2 rmind KASSERT(error == 0); 159 1.5.2.2 rmind } 160 1.5.2.2 rmind 161 1.5.2.2 rmind /* 162 1.5.2.2 rmind * Allocate a new, uninitialized vnode. If 'mp' is non-NULL, this is a 163 1.5.2.2 rmind * marker vnode and we are prepared to wait for the allocation. 164 1.5.2.2 rmind */ 165 1.5.2.2 rmind vnode_t * 166 1.5.2.2 rmind vnalloc(struct mount *mp) 167 1.5.2.2 rmind { 168 1.5.2.2 rmind vnode_t *vp; 169 1.5.2.2 rmind 170 1.5.2.2 rmind vp = pool_cache_get(vnode_cache, (mp != NULL ? PR_WAITOK : PR_NOWAIT)); 171 1.5.2.2 rmind if (vp == NULL) { 172 1.5.2.2 rmind return NULL; 173 1.5.2.2 rmind } 174 1.5.2.2 rmind 175 1.5.2.2 rmind memset(vp, 0, sizeof(*vp)); 176 1.5.2.2 rmind uvm_obj_init(&vp->v_uobj, &uvm_vnodeops, NULL, 0); 177 1.5.2.2 rmind cv_init(&vp->v_cv, "vnode"); 178 1.5.2.2 rmind /* 179 1.5.2.2 rmind * Done by memset() above. 180 1.5.2.2 rmind * LIST_INIT(&vp->v_nclist); 181 1.5.2.2 rmind * LIST_INIT(&vp->v_dnclist); 182 1.5.2.2 rmind */ 183 1.5.2.2 rmind 184 1.5.2.2 rmind if (mp != NULL) { 185 1.5.2.2 rmind vp->v_mount = mp; 186 1.5.2.2 rmind vp->v_type = VBAD; 187 1.5.2.2 rmind vp->v_iflag = VI_MARKER; 188 1.5.2.2 rmind } else { 189 1.5.2.2 rmind rw_init(&vp->v_lock); 190 1.5.2.2 rmind } 191 1.5.2.2 rmind 192 1.5.2.2 rmind return vp; 193 1.5.2.2 rmind } 194 1.5.2.2 rmind 195 1.5.2.2 rmind /* 196 1.5.2.2 rmind * Free an unused, unreferenced vnode. 197 1.5.2.2 rmind */ 198 1.5.2.2 rmind void 199 1.5.2.2 rmind vnfree(vnode_t *vp) 200 1.5.2.2 rmind { 201 1.5.2.2 rmind 202 1.5.2.2 rmind KASSERT(vp->v_usecount == 0); 203 1.5.2.2 rmind 204 1.5.2.2 rmind if ((vp->v_iflag & VI_MARKER) == 0) { 205 1.5.2.2 rmind rw_destroy(&vp->v_lock); 206 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 207 1.5.2.2 rmind numvnodes--; 208 1.5.2.2 rmind mutex_exit(&vnode_free_list_lock); 209 1.5.2.2 rmind } 210 1.5.2.2 rmind 211 1.5.2.2 rmind uvm_obj_destroy(&vp->v_uobj, NULL); 212 1.5.2.2 rmind cv_destroy(&vp->v_cv); 213 1.5.2.2 rmind pool_cache_put(vnode_cache, vp); 214 1.5.2.2 rmind } 215 1.5.2.2 rmind 216 1.5.2.2 rmind /* 217 1.5.2.2 rmind * getcleanvnode: grab a vnode from freelist and clean it. 218 1.5.2.2 rmind * 219 1.5.2.2 rmind * => Releases vnode_free_list_lock. 220 1.5.2.2 rmind * => Returns referenced vnode on success. 221 1.5.2.2 rmind */ 222 1.5.2.2 rmind static vnode_t * 223 1.5.2.2 rmind getcleanvnode(void) 224 1.5.2.2 rmind { 225 1.5.2.2 rmind vnode_t *vp; 226 1.5.2.2 rmind vnodelst_t *listhd; 227 1.5.2.2 rmind 228 1.5.2.2 rmind KASSERT(mutex_owned(&vnode_free_list_lock)); 229 1.5.2.2 rmind retry: 230 1.5.2.2 rmind listhd = &vnode_free_list; 231 1.5.2.2 rmind try_nextlist: 232 1.5.2.2 rmind TAILQ_FOREACH(vp, listhd, v_freelist) { 233 1.5.2.2 rmind /* 234 1.5.2.2 rmind * It's safe to test v_usecount and v_iflag 235 1.5.2.2 rmind * without holding the interlock here, since 236 1.5.2.2 rmind * these vnodes should never appear on the 237 1.5.2.2 rmind * lists. 238 1.5.2.2 rmind */ 239 1.5.2.2 rmind KASSERT(vp->v_usecount == 0); 240 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_CLEAN) == 0); 241 1.5.2.2 rmind KASSERT(vp->v_freelisthd == listhd); 242 1.5.2.2 rmind 243 1.5.2.2 rmind if (!mutex_tryenter(vp->v_interlock)) 244 1.5.2.2 rmind continue; 245 1.5.2.2 rmind if ((vp->v_iflag & VI_XLOCK) == 0) 246 1.5.2.2 rmind break; 247 1.5.2.2 rmind mutex_exit(vp->v_interlock); 248 1.5.2.2 rmind } 249 1.5.2.2 rmind 250 1.5.2.2 rmind if (vp == NULL) { 251 1.5.2.2 rmind if (listhd == &vnode_free_list) { 252 1.5.2.2 rmind listhd = &vnode_hold_list; 253 1.5.2.2 rmind goto try_nextlist; 254 1.5.2.2 rmind } 255 1.5.2.2 rmind mutex_exit(&vnode_free_list_lock); 256 1.5.2.2 rmind return NULL; 257 1.5.2.2 rmind } 258 1.5.2.2 rmind 259 1.5.2.2 rmind /* Remove it from the freelist. */ 260 1.5.2.2 rmind TAILQ_REMOVE(listhd, vp, v_freelist); 261 1.5.2.2 rmind vp->v_freelisthd = NULL; 262 1.5.2.2 rmind mutex_exit(&vnode_free_list_lock); 263 1.5.2.2 rmind 264 1.5.2.2 rmind KASSERT(vp->v_usecount == 0); 265 1.5.2.2 rmind 266 1.5.2.2 rmind /* 267 1.5.2.2 rmind * The vnode is still associated with a file system, so we must 268 1.5.2.2 rmind * clean it out before reusing it. We need to add a reference 269 1.5.2.2 rmind * before doing this. If the vnode gains another reference while 270 1.5.2.2 rmind * being cleaned out then we lose - retry. 271 1.5.2.2 rmind */ 272 1.5.2.2 rmind atomic_add_int(&vp->v_usecount, 1 + VC_XLOCK); 273 1.5.2.2 rmind vclean(vp, DOCLOSE); 274 1.5.2.2 rmind KASSERT(vp->v_usecount >= 1 + VC_XLOCK); 275 1.5.2.2 rmind atomic_add_int(&vp->v_usecount, -VC_XLOCK); 276 1.5.2.2 rmind if (vp->v_usecount == 1) { 277 1.5.2.2 rmind /* We're about to dirty it. */ 278 1.5.2.2 rmind vp->v_iflag &= ~VI_CLEAN; 279 1.5.2.2 rmind mutex_exit(vp->v_interlock); 280 1.5.2.2 rmind if (vp->v_type == VBLK || vp->v_type == VCHR) { 281 1.5.2.2 rmind spec_node_destroy(vp); 282 1.5.2.2 rmind } 283 1.5.2.2 rmind vp->v_type = VNON; 284 1.5.2.2 rmind } else { 285 1.5.2.2 rmind /* 286 1.5.2.2 rmind * Don't return to freelist - the holder of the last 287 1.5.2.2 rmind * reference will destroy it. 288 1.5.2.2 rmind */ 289 1.5.2.2 rmind vrelel(vp, 0); /* releases vp->v_interlock */ 290 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 291 1.5.2.2 rmind goto retry; 292 1.5.2.2 rmind } 293 1.5.2.2 rmind 294 1.5.2.2 rmind KASSERT(vp->v_data == NULL); 295 1.5.2.2 rmind KASSERT(vp->v_uobj.uo_npages == 0); 296 1.5.2.2 rmind KASSERT(TAILQ_EMPTY(&vp->v_uobj.memq)); 297 1.5.2.2 rmind KASSERT(vp->v_numoutput == 0); 298 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_ONWORKLST) == 0); 299 1.5.2.2 rmind 300 1.5.2.2 rmind return vp; 301 1.5.2.2 rmind } 302 1.5.2.2 rmind 303 1.5.2.2 rmind /* 304 1.5.2.2 rmind * getnewvnode: return the next vnode from the free list. 305 1.5.2.2 rmind * 306 1.5.2.2 rmind * => Returns referenced vnode, moved into the mount queue. 307 1.5.2.2 rmind */ 308 1.5.2.2 rmind int 309 1.5.2.2 rmind getnewvnode(enum vtagtype tag, struct mount *mp, int (**vops)(void *), 310 1.5.2.2 rmind vnode_t **vpp) 311 1.5.2.2 rmind { 312 1.5.2.2 rmind struct uvm_object *uobj; 313 1.5.2.2 rmind static int toggle; 314 1.5.2.2 rmind vnode_t *vp; 315 1.5.2.2 rmind int error = 0, tryalloc; 316 1.5.2.2 rmind 317 1.5.2.2 rmind try_again: 318 1.5.2.2 rmind if (mp != NULL) { 319 1.5.2.2 rmind /* 320 1.5.2.2 rmind * Mark filesystem busy while we are creating a vnode. 321 1.5.2.2 rmind * If unmount is in progress, this will fail. 322 1.5.2.2 rmind */ 323 1.5.2.2 rmind error = vfs_busy(mp, NULL); 324 1.5.2.2 rmind if (error) 325 1.5.2.2 rmind return error; 326 1.5.2.2 rmind } 327 1.5.2.2 rmind 328 1.5.2.2 rmind /* 329 1.5.2.2 rmind * We must choose whether to allocate a new vnode or recycle an 330 1.5.2.2 rmind * existing one. The criterion for allocating a new one is that 331 1.5.2.2 rmind * the total number of vnodes is less than the number desired or 332 1.5.2.2 rmind * there are no vnodes on either free list. Generally we only 333 1.5.2.2 rmind * want to recycle vnodes that have no buffers associated with 334 1.5.2.2 rmind * them, so we look first on the vnode_free_list. If it is empty, 335 1.5.2.2 rmind * we next consider vnodes with referencing buffers on the 336 1.5.2.2 rmind * vnode_hold_list. The toggle ensures that half the time we 337 1.5.2.2 rmind * will use a buffer from the vnode_hold_list, and half the time 338 1.5.2.2 rmind * we will allocate a new one unless the list has grown to twice 339 1.5.2.2 rmind * the desired size. We are reticent to recycle vnodes from the 340 1.5.2.2 rmind * vnode_hold_list because we will lose the identity of all its 341 1.5.2.2 rmind * referencing buffers. 342 1.5.2.2 rmind */ 343 1.5.2.2 rmind 344 1.5.2.2 rmind vp = NULL; 345 1.5.2.2 rmind 346 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 347 1.5.2.2 rmind 348 1.5.2.2 rmind toggle ^= 1; 349 1.5.2.2 rmind if (numvnodes > 2 * desiredvnodes) 350 1.5.2.2 rmind toggle = 0; 351 1.5.2.2 rmind 352 1.5.2.2 rmind tryalloc = numvnodes < desiredvnodes || 353 1.5.2.2 rmind (TAILQ_FIRST(&vnode_free_list) == NULL && 354 1.5.2.2 rmind (TAILQ_FIRST(&vnode_hold_list) == NULL || toggle)); 355 1.5.2.2 rmind 356 1.5.2.2 rmind if (tryalloc) { 357 1.5.2.2 rmind /* Allocate a new vnode. */ 358 1.5.2.2 rmind numvnodes++; 359 1.5.2.2 rmind mutex_exit(&vnode_free_list_lock); 360 1.5.2.2 rmind if ((vp = vnalloc(NULL)) == NULL) { 361 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 362 1.5.2.2 rmind numvnodes--; 363 1.5.2.2 rmind } else 364 1.5.2.2 rmind vp->v_usecount = 1; 365 1.5.2.2 rmind } 366 1.5.2.2 rmind 367 1.5.2.2 rmind if (vp == NULL) { 368 1.5.2.2 rmind /* Recycle and get vnode clean. */ 369 1.5.2.2 rmind vp = getcleanvnode(); 370 1.5.2.2 rmind if (vp == NULL) { 371 1.5.2.2 rmind if (mp != NULL) { 372 1.5.2.2 rmind vfs_unbusy(mp, false, NULL); 373 1.5.2.2 rmind } 374 1.5.2.2 rmind if (tryalloc) { 375 1.5.2.2 rmind printf("WARNING: unable to allocate new " 376 1.5.2.2 rmind "vnode, retrying...\n"); 377 1.5.2.2 rmind kpause("newvn", false, hz, NULL); 378 1.5.2.2 rmind goto try_again; 379 1.5.2.2 rmind } 380 1.5.2.2 rmind tablefull("vnode", "increase kern.maxvnodes or NVNODE"); 381 1.5.2.2 rmind *vpp = 0; 382 1.5.2.2 rmind return ENFILE; 383 1.5.2.2 rmind } 384 1.5.2.2 rmind vp->v_iflag = 0; 385 1.5.2.2 rmind vp->v_vflag = 0; 386 1.5.2.2 rmind vp->v_uflag = 0; 387 1.5.2.2 rmind vp->v_socket = NULL; 388 1.5.2.2 rmind } 389 1.5.2.2 rmind 390 1.5.2.2 rmind KASSERT(vp->v_usecount == 1); 391 1.5.2.2 rmind KASSERT(vp->v_freelisthd == NULL); 392 1.5.2.2 rmind KASSERT(LIST_EMPTY(&vp->v_nclist)); 393 1.5.2.2 rmind KASSERT(LIST_EMPTY(&vp->v_dnclist)); 394 1.5.2.2 rmind 395 1.5.2.2 rmind /* Initialize vnode. */ 396 1.5.2.2 rmind vp->v_type = VNON; 397 1.5.2.2 rmind vp->v_tag = tag; 398 1.5.2.2 rmind vp->v_op = vops; 399 1.5.2.2 rmind vp->v_data = NULL; 400 1.5.2.2 rmind 401 1.5.2.2 rmind uobj = &vp->v_uobj; 402 1.5.2.2 rmind KASSERT(uobj->pgops == &uvm_vnodeops); 403 1.5.2.2 rmind KASSERT(uobj->uo_npages == 0); 404 1.5.2.2 rmind KASSERT(TAILQ_FIRST(&uobj->memq) == NULL); 405 1.5.2.2 rmind vp->v_size = vp->v_writesize = VSIZENOTSET; 406 1.5.2.2 rmind 407 1.5.2.2 rmind /* Finally, move vnode into the mount queue. */ 408 1.5.2.2 rmind vfs_insmntque(vp, mp); 409 1.5.2.2 rmind 410 1.5.2.2 rmind if (mp != NULL) { 411 1.5.2.2 rmind if ((mp->mnt_iflag & IMNT_MPSAFE) != 0) 412 1.5.2.2 rmind vp->v_vflag |= VV_MPSAFE; 413 1.5.2.2 rmind vfs_unbusy(mp, true, NULL); 414 1.5.2.2 rmind } 415 1.5.2.2 rmind 416 1.5.2.2 rmind *vpp = vp; 417 1.5.2.2 rmind return 0; 418 1.5.2.2 rmind } 419 1.5.2.2 rmind 420 1.5.2.2 rmind /* 421 1.5.2.2 rmind * This is really just the reverse of getnewvnode(). Needed for 422 1.5.2.2 rmind * VFS_VGET functions who may need to push back a vnode in case 423 1.5.2.2 rmind * of a locking race. 424 1.5.2.2 rmind */ 425 1.5.2.2 rmind void 426 1.5.2.2 rmind ungetnewvnode(vnode_t *vp) 427 1.5.2.2 rmind { 428 1.5.2.2 rmind 429 1.5.2.2 rmind KASSERT(vp->v_usecount == 1); 430 1.5.2.2 rmind KASSERT(vp->v_data == NULL); 431 1.5.2.2 rmind KASSERT(vp->v_freelisthd == NULL); 432 1.5.2.2 rmind 433 1.5.2.2 rmind mutex_enter(vp->v_interlock); 434 1.5.2.2 rmind vp->v_iflag |= VI_CLEAN; 435 1.5.2.2 rmind vrelel(vp, 0); 436 1.5.2.2 rmind } 437 1.5.2.2 rmind 438 1.5.2.2 rmind /* 439 1.5.2.2 rmind * Remove a vnode from its freelist. 440 1.5.2.2 rmind */ 441 1.5.2.2 rmind void 442 1.5.2.2 rmind vremfree(vnode_t *vp) 443 1.5.2.2 rmind { 444 1.5.2.2 rmind 445 1.5.2.2 rmind KASSERT(mutex_owned(vp->v_interlock)); 446 1.5.2.2 rmind KASSERT(vp->v_usecount == 0); 447 1.5.2.2 rmind 448 1.5.2.2 rmind /* 449 1.5.2.2 rmind * Note that the reference count must not change until 450 1.5.2.2 rmind * the vnode is removed. 451 1.5.2.2 rmind */ 452 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 453 1.5.2.2 rmind if (vp->v_holdcnt > 0) { 454 1.5.2.2 rmind KASSERT(vp->v_freelisthd == &vnode_hold_list); 455 1.5.2.2 rmind } else { 456 1.5.2.2 rmind KASSERT(vp->v_freelisthd == &vnode_free_list); 457 1.5.2.2 rmind } 458 1.5.2.2 rmind TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 459 1.5.2.2 rmind vp->v_freelisthd = NULL; 460 1.5.2.2 rmind mutex_exit(&vnode_free_list_lock); 461 1.5.2.2 rmind } 462 1.5.2.2 rmind 463 1.5.2.2 rmind /* 464 1.5.2.2 rmind * Try to gain a reference to a vnode, without acquiring its interlock. 465 1.5.2.2 rmind * The caller must hold a lock that will prevent the vnode from being 466 1.5.2.2 rmind * recycled or freed. 467 1.5.2.2 rmind */ 468 1.5.2.2 rmind bool 469 1.5.2.2 rmind vtryget(vnode_t *vp) 470 1.5.2.2 rmind { 471 1.5.2.2 rmind u_int use, next; 472 1.5.2.2 rmind 473 1.5.2.2 rmind /* 474 1.5.2.2 rmind * If the vnode is being freed, don't make life any harder 475 1.5.2.2 rmind * for vclean() by adding another reference without waiting. 476 1.5.2.2 rmind * This is not strictly necessary, but we'll do it anyway. 477 1.5.2.2 rmind */ 478 1.5.2.2 rmind if (__predict_false((vp->v_iflag & VI_XLOCK) != 0)) { 479 1.5.2.2 rmind return false; 480 1.5.2.2 rmind } 481 1.5.2.2 rmind for (use = vp->v_usecount;; use = next) { 482 1.5.2.2 rmind if (use == 0 || __predict_false((use & VC_XLOCK) != 0)) { 483 1.5.2.2 rmind /* Need interlock held if first reference. */ 484 1.5.2.2 rmind return false; 485 1.5.2.2 rmind } 486 1.5.2.2 rmind next = atomic_cas_uint(&vp->v_usecount, use, use + 1); 487 1.5.2.2 rmind if (__predict_true(next == use)) { 488 1.5.2.2 rmind return true; 489 1.5.2.2 rmind } 490 1.5.2.2 rmind } 491 1.5.2.2 rmind } 492 1.5.2.2 rmind 493 1.5.2.2 rmind /* 494 1.5.2.2 rmind * vget: get a particular vnode from the free list, increment its reference 495 1.5.2.2 rmind * count and lock it. 496 1.5.2.2 rmind * 497 1.5.2.2 rmind * => Should be called with v_interlock held. 498 1.5.2.2 rmind * 499 1.5.2.2 rmind * If VI_XLOCK is set, the vnode is being eliminated in vgone()/vclean(). 500 1.5.2.2 rmind * In that case, we cannot grab the vnode, so the process is awakened when 501 1.5.2.2 rmind * the transition is completed, and an error returned to indicate that the 502 1.5.2.2 rmind * vnode is no longer usable (e.g. changed to a new file system type). 503 1.5.2.2 rmind */ 504 1.5.2.2 rmind int 505 1.5.2.2 rmind vget(vnode_t *vp, int flags) 506 1.5.2.2 rmind { 507 1.5.2.2 rmind int error = 0; 508 1.5.2.2 rmind 509 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 510 1.5.2.2 rmind KASSERT(mutex_owned(vp->v_interlock)); 511 1.5.2.2 rmind KASSERT((flags & ~(LK_SHARED|LK_EXCLUSIVE|LK_NOWAIT)) == 0); 512 1.5.2.2 rmind 513 1.5.2.2 rmind /* 514 1.5.2.2 rmind * Before adding a reference, we must remove the vnode 515 1.5.2.2 rmind * from its freelist. 516 1.5.2.2 rmind */ 517 1.5.2.2 rmind if (vp->v_usecount == 0) { 518 1.5.2.2 rmind vremfree(vp); 519 1.5.2.2 rmind vp->v_usecount = 1; 520 1.5.2.2 rmind } else { 521 1.5.2.2 rmind atomic_inc_uint(&vp->v_usecount); 522 1.5.2.2 rmind } 523 1.5.2.2 rmind 524 1.5.2.2 rmind /* 525 1.5.2.2 rmind * If the vnode is in the process of being cleaned out for 526 1.5.2.2 rmind * another use, we wait for the cleaning to finish and then 527 1.5.2.2 rmind * return failure. Cleaning is determined by checking if 528 1.5.2.2 rmind * the VI_XLOCK flag is set. 529 1.5.2.2 rmind */ 530 1.5.2.2 rmind if ((vp->v_iflag & VI_XLOCK) != 0) { 531 1.5.2.2 rmind if ((flags & LK_NOWAIT) != 0) { 532 1.5.2.2 rmind vrelel(vp, 0); 533 1.5.2.2 rmind return EBUSY; 534 1.5.2.2 rmind } 535 1.5.2.2 rmind vwait(vp, VI_XLOCK); 536 1.5.2.2 rmind vrelel(vp, 0); 537 1.5.2.2 rmind return ENOENT; 538 1.5.2.2 rmind } 539 1.5.2.2 rmind 540 1.5.2.2 rmind /* 541 1.5.2.2 rmind * Ok, we got it in good shape. Just locking left. 542 1.5.2.2 rmind */ 543 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_CLEAN) == 0); 544 1.5.2.2 rmind mutex_exit(vp->v_interlock); 545 1.5.2.2 rmind if (flags & (LK_EXCLUSIVE | LK_SHARED)) { 546 1.5.2.2 rmind error = vn_lock(vp, flags); 547 1.5.2.2 rmind if (error != 0) { 548 1.5.2.2 rmind vrele(vp); 549 1.5.2.2 rmind } 550 1.5.2.2 rmind } 551 1.5.2.2 rmind return error; 552 1.5.2.2 rmind } 553 1.5.2.2 rmind 554 1.5.2.2 rmind /* 555 1.5.2.2 rmind * vput: unlock and release the reference. 556 1.5.2.2 rmind */ 557 1.5.2.2 rmind void 558 1.5.2.2 rmind vput(vnode_t *vp) 559 1.5.2.2 rmind { 560 1.5.2.2 rmind 561 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 562 1.5.2.2 rmind 563 1.5.2.2 rmind VOP_UNLOCK(vp); 564 1.5.2.2 rmind vrele(vp); 565 1.5.2.2 rmind } 566 1.5.2.2 rmind 567 1.5.2.2 rmind /* 568 1.5.2.2 rmind * Try to drop reference on a vnode. Abort if we are releasing the 569 1.5.2.2 rmind * last reference. Note: this _must_ succeed if not the last reference. 570 1.5.2.2 rmind */ 571 1.5.2.2 rmind static inline bool 572 1.5.2.2 rmind vtryrele(vnode_t *vp) 573 1.5.2.2 rmind { 574 1.5.2.2 rmind u_int use, next; 575 1.5.2.2 rmind 576 1.5.2.2 rmind for (use = vp->v_usecount;; use = next) { 577 1.5.2.2 rmind if (use == 1) { 578 1.5.2.2 rmind return false; 579 1.5.2.2 rmind } 580 1.5.2.2 rmind KASSERT((use & VC_MASK) > 1); 581 1.5.2.2 rmind next = atomic_cas_uint(&vp->v_usecount, use, use - 1); 582 1.5.2.2 rmind if (__predict_true(next == use)) { 583 1.5.2.2 rmind return true; 584 1.5.2.2 rmind } 585 1.5.2.2 rmind } 586 1.5.2.2 rmind } 587 1.5.2.2 rmind 588 1.5.2.2 rmind /* 589 1.5.2.2 rmind * Vnode release. If reference count drops to zero, call inactive 590 1.5.2.2 rmind * routine and either return to freelist or free to the pool. 591 1.5.2.2 rmind */ 592 1.5.2.2 rmind void 593 1.5.2.2 rmind vrelel(vnode_t *vp, int flags) 594 1.5.2.2 rmind { 595 1.5.2.2 rmind bool recycle, defer; 596 1.5.2.2 rmind int error; 597 1.5.2.2 rmind 598 1.5.2.2 rmind KASSERT(mutex_owned(vp->v_interlock)); 599 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 600 1.5.2.2 rmind KASSERT(vp->v_freelisthd == NULL); 601 1.5.2.2 rmind 602 1.5.2.2 rmind if (__predict_false(vp->v_op == dead_vnodeop_p && 603 1.5.2.2 rmind (vp->v_iflag & (VI_CLEAN|VI_XLOCK)) == 0)) { 604 1.5.2.2 rmind vpanic(vp, "dead but not clean"); 605 1.5.2.2 rmind } 606 1.5.2.2 rmind 607 1.5.2.2 rmind /* 608 1.5.2.2 rmind * If not the last reference, just drop the reference count 609 1.5.2.2 rmind * and unlock. 610 1.5.2.2 rmind */ 611 1.5.2.2 rmind if (vtryrele(vp)) { 612 1.5.2.2 rmind vp->v_iflag |= VI_INACTREDO; 613 1.5.2.2 rmind mutex_exit(vp->v_interlock); 614 1.5.2.2 rmind return; 615 1.5.2.2 rmind } 616 1.5.2.2 rmind if (vp->v_usecount <= 0 || vp->v_writecount != 0) { 617 1.5.2.2 rmind vpanic(vp, "vrelel: bad ref count"); 618 1.5.2.2 rmind } 619 1.5.2.2 rmind 620 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_XLOCK) == 0); 621 1.5.2.2 rmind 622 1.5.2.2 rmind /* 623 1.5.2.2 rmind * If not clean, deactivate the vnode, but preserve 624 1.5.2.2 rmind * our reference across the call to VOP_INACTIVE(). 625 1.5.2.2 rmind */ 626 1.5.2.2 rmind retry: 627 1.5.2.2 rmind if ((vp->v_iflag & VI_CLEAN) == 0) { 628 1.5.2.2 rmind recycle = false; 629 1.5.2.2 rmind vp->v_iflag |= VI_INACTNOW; 630 1.5.2.2 rmind 631 1.5.2.2 rmind /* 632 1.5.2.2 rmind * XXX This ugly block can be largely eliminated if 633 1.5.2.2 rmind * locking is pushed down into the file systems. 634 1.5.2.2 rmind * 635 1.5.2.2 rmind * Defer vnode release to vrele_thread if caller 636 1.5.2.2 rmind * requests it explicitly. 637 1.5.2.2 rmind */ 638 1.5.2.2 rmind if ((curlwp == uvm.pagedaemon_lwp) || 639 1.5.2.2 rmind (flags & VRELEL_ASYNC_RELE) != 0) { 640 1.5.2.2 rmind /* The pagedaemon can't wait around; defer. */ 641 1.5.2.2 rmind defer = true; 642 1.5.2.2 rmind } else if (curlwp == vrele_lwp) { 643 1.5.2.2 rmind /* We have to try harder. */ 644 1.5.2.2 rmind vp->v_iflag &= ~VI_INACTREDO; 645 1.5.2.2 rmind mutex_exit(vp->v_interlock); 646 1.5.2.2 rmind error = vn_lock(vp, LK_EXCLUSIVE); 647 1.5.2.2 rmind if (error != 0) { 648 1.5.2.2 rmind /* XXX */ 649 1.5.2.2 rmind vpanic(vp, "vrele: unable to lock %p"); 650 1.5.2.2 rmind } 651 1.5.2.2 rmind defer = false; 652 1.5.2.2 rmind } else if ((vp->v_iflag & VI_LAYER) != 0) { 653 1.5.2.2 rmind /* 654 1.5.2.2 rmind * Acquiring the stack's lock in vclean() even 655 1.5.2.2 rmind * for an honest vput/vrele is dangerous because 656 1.5.2.2 rmind * our caller may hold other vnode locks; defer. 657 1.5.2.2 rmind */ 658 1.5.2.2 rmind defer = true; 659 1.5.2.2 rmind } else { 660 1.5.2.2 rmind /* If we can't acquire the lock, then defer. */ 661 1.5.2.2 rmind vp->v_iflag &= ~VI_INACTREDO; 662 1.5.2.2 rmind mutex_exit(vp->v_interlock); 663 1.5.2.2 rmind error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT); 664 1.5.2.2 rmind if (error != 0) { 665 1.5.2.2 rmind defer = true; 666 1.5.2.2 rmind mutex_enter(vp->v_interlock); 667 1.5.2.2 rmind } else { 668 1.5.2.2 rmind defer = false; 669 1.5.2.2 rmind } 670 1.5.2.2 rmind } 671 1.5.2.2 rmind 672 1.5.2.2 rmind if (defer) { 673 1.5.2.2 rmind /* 674 1.5.2.2 rmind * Defer reclaim to the kthread; it's not safe to 675 1.5.2.2 rmind * clean it here. We donate it our last reference. 676 1.5.2.2 rmind */ 677 1.5.2.2 rmind KASSERT(mutex_owned(vp->v_interlock)); 678 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_INACTPEND) == 0); 679 1.5.2.2 rmind vp->v_iflag &= ~VI_INACTNOW; 680 1.5.2.2 rmind vp->v_iflag |= VI_INACTPEND; 681 1.5.2.2 rmind mutex_enter(&vrele_lock); 682 1.5.2.2 rmind TAILQ_INSERT_TAIL(&vrele_list, vp, v_freelist); 683 1.5.2.2 rmind if (++vrele_pending > (desiredvnodes >> 8)) 684 1.5.2.2 rmind cv_signal(&vrele_cv); 685 1.5.2.2 rmind mutex_exit(&vrele_lock); 686 1.5.2.2 rmind mutex_exit(vp->v_interlock); 687 1.5.2.2 rmind return; 688 1.5.2.2 rmind } 689 1.5.2.2 rmind 690 1.5.2.2 rmind #ifdef DIAGNOSTIC 691 1.5.2.2 rmind if ((vp->v_type == VBLK || vp->v_type == VCHR) && 692 1.5.2.2 rmind vp->v_specnode != NULL && vp->v_specnode->sn_opencnt != 0) { 693 1.5.2.2 rmind vprint("vrelel: missing VOP_CLOSE()", vp); 694 1.5.2.2 rmind } 695 1.5.2.2 rmind #endif 696 1.5.2.2 rmind 697 1.5.2.2 rmind /* 698 1.5.2.2 rmind * The vnode can gain another reference while being 699 1.5.2.2 rmind * deactivated. If VOP_INACTIVE() indicates that 700 1.5.2.2 rmind * the described file has been deleted, then recycle 701 1.5.2.2 rmind * the vnode irrespective of additional references. 702 1.5.2.2 rmind * Another thread may be waiting to re-use the on-disk 703 1.5.2.2 rmind * inode. 704 1.5.2.2 rmind * 705 1.5.2.2 rmind * Note that VOP_INACTIVE() will drop the vnode lock. 706 1.5.2.2 rmind */ 707 1.5.2.2 rmind VOP_INACTIVE(vp, &recycle); 708 1.5.2.2 rmind mutex_enter(vp->v_interlock); 709 1.5.2.2 rmind vp->v_iflag &= ~VI_INACTNOW; 710 1.5.2.2 rmind if (!recycle) { 711 1.5.2.2 rmind if (vtryrele(vp)) { 712 1.5.2.2 rmind mutex_exit(vp->v_interlock); 713 1.5.2.2 rmind return; 714 1.5.2.2 rmind } 715 1.5.2.2 rmind 716 1.5.2.2 rmind /* 717 1.5.2.2 rmind * If we grew another reference while 718 1.5.2.2 rmind * VOP_INACTIVE() was underway, retry. 719 1.5.2.2 rmind */ 720 1.5.2.2 rmind if ((vp->v_iflag & VI_INACTREDO) != 0) { 721 1.5.2.2 rmind goto retry; 722 1.5.2.2 rmind } 723 1.5.2.2 rmind } 724 1.5.2.2 rmind 725 1.5.2.2 rmind /* Take care of space accounting. */ 726 1.5.2.2 rmind if (vp->v_iflag & VI_EXECMAP) { 727 1.5.2.2 rmind atomic_add_int(&uvmexp.execpages, 728 1.5.2.2 rmind -vp->v_uobj.uo_npages); 729 1.5.2.2 rmind atomic_add_int(&uvmexp.filepages, 730 1.5.2.2 rmind vp->v_uobj.uo_npages); 731 1.5.2.2 rmind } 732 1.5.2.2 rmind vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP|VI_WRMAP); 733 1.5.2.2 rmind vp->v_vflag &= ~VV_MAPPED; 734 1.5.2.2 rmind 735 1.5.2.2 rmind /* 736 1.5.2.2 rmind * Recycle the vnode if the file is now unused (unlinked), 737 1.5.2.2 rmind * otherwise just free it. 738 1.5.2.2 rmind */ 739 1.5.2.2 rmind if (recycle) { 740 1.5.2.2 rmind vclean(vp, DOCLOSE); 741 1.5.2.2 rmind } 742 1.5.2.2 rmind KASSERT(vp->v_usecount > 0); 743 1.5.2.2 rmind } 744 1.5.2.2 rmind 745 1.5.2.2 rmind if (atomic_dec_uint_nv(&vp->v_usecount) != 0) { 746 1.5.2.2 rmind /* Gained another reference while being reclaimed. */ 747 1.5.2.2 rmind mutex_exit(vp->v_interlock); 748 1.5.2.2 rmind return; 749 1.5.2.2 rmind } 750 1.5.2.2 rmind 751 1.5.2.2 rmind if ((vp->v_iflag & VI_CLEAN) != 0) { 752 1.5.2.2 rmind /* 753 1.5.2.2 rmind * It's clean so destroy it. It isn't referenced 754 1.5.2.2 rmind * anywhere since it has been reclaimed. 755 1.5.2.2 rmind */ 756 1.5.2.2 rmind KASSERT(vp->v_holdcnt == 0); 757 1.5.2.2 rmind KASSERT(vp->v_writecount == 0); 758 1.5.2.2 rmind mutex_exit(vp->v_interlock); 759 1.5.2.2 rmind vfs_insmntque(vp, NULL); 760 1.5.2.2 rmind if (vp->v_type == VBLK || vp->v_type == VCHR) { 761 1.5.2.2 rmind spec_node_destroy(vp); 762 1.5.2.2 rmind } 763 1.5.2.2 rmind vnfree(vp); 764 1.5.2.2 rmind } else { 765 1.5.2.2 rmind /* 766 1.5.2.2 rmind * Otherwise, put it back onto the freelist. It 767 1.5.2.2 rmind * can't be destroyed while still associated with 768 1.5.2.2 rmind * a file system. 769 1.5.2.2 rmind */ 770 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 771 1.5.2.2 rmind if (vp->v_holdcnt > 0) { 772 1.5.2.2 rmind vp->v_freelisthd = &vnode_hold_list; 773 1.5.2.2 rmind } else { 774 1.5.2.2 rmind vp->v_freelisthd = &vnode_free_list; 775 1.5.2.2 rmind } 776 1.5.2.2 rmind TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 777 1.5.2.2 rmind mutex_exit(&vnode_free_list_lock); 778 1.5.2.2 rmind mutex_exit(vp->v_interlock); 779 1.5.2.2 rmind } 780 1.5.2.2 rmind } 781 1.5.2.2 rmind 782 1.5.2.2 rmind void 783 1.5.2.2 rmind vrele(vnode_t *vp) 784 1.5.2.2 rmind { 785 1.5.2.2 rmind 786 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 787 1.5.2.2 rmind 788 1.5.2.2 rmind if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) { 789 1.5.2.2 rmind return; 790 1.5.2.2 rmind } 791 1.5.2.2 rmind mutex_enter(vp->v_interlock); 792 1.5.2.2 rmind vrelel(vp, 0); 793 1.5.2.2 rmind } 794 1.5.2.2 rmind 795 1.5.2.2 rmind /* 796 1.5.2.2 rmind * Asynchronous vnode release, vnode is released in different context. 797 1.5.2.2 rmind */ 798 1.5.2.2 rmind void 799 1.5.2.2 rmind vrele_async(vnode_t *vp) 800 1.5.2.2 rmind { 801 1.5.2.2 rmind 802 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 803 1.5.2.2 rmind 804 1.5.2.2 rmind if ((vp->v_iflag & VI_INACTNOW) == 0 && vtryrele(vp)) { 805 1.5.2.2 rmind return; 806 1.5.2.2 rmind } 807 1.5.2.2 rmind mutex_enter(vp->v_interlock); 808 1.5.2.2 rmind vrelel(vp, VRELEL_ASYNC_RELE); 809 1.5.2.2 rmind } 810 1.5.2.2 rmind 811 1.5.2.2 rmind static void 812 1.5.2.2 rmind vrele_thread(void *cookie) 813 1.5.2.2 rmind { 814 1.5.2.2 rmind vnode_t *vp; 815 1.5.2.2 rmind 816 1.5.2.2 rmind for (;;) { 817 1.5.2.2 rmind mutex_enter(&vrele_lock); 818 1.5.2.2 rmind while (TAILQ_EMPTY(&vrele_list)) { 819 1.5.2.2 rmind vrele_gen++; 820 1.5.2.2 rmind cv_broadcast(&vrele_cv); 821 1.5.2.2 rmind cv_timedwait(&vrele_cv, &vrele_lock, hz); 822 1.5.2.2 rmind } 823 1.5.2.2 rmind vp = TAILQ_FIRST(&vrele_list); 824 1.5.2.2 rmind TAILQ_REMOVE(&vrele_list, vp, v_freelist); 825 1.5.2.2 rmind vrele_pending--; 826 1.5.2.2 rmind mutex_exit(&vrele_lock); 827 1.5.2.2 rmind 828 1.5.2.2 rmind /* 829 1.5.2.2 rmind * If not the last reference, then ignore the vnode 830 1.5.2.2 rmind * and look for more work. 831 1.5.2.2 rmind */ 832 1.5.2.2 rmind mutex_enter(vp->v_interlock); 833 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_INACTPEND) != 0); 834 1.5.2.2 rmind vp->v_iflag &= ~VI_INACTPEND; 835 1.5.2.2 rmind vrelel(vp, 0); 836 1.5.2.2 rmind } 837 1.5.2.2 rmind } 838 1.5.2.2 rmind 839 1.5.2.2 rmind void 840 1.5.2.2 rmind vrele_flush(void) 841 1.5.2.2 rmind { 842 1.5.2.2 rmind int gen; 843 1.5.2.2 rmind 844 1.5.2.2 rmind mutex_enter(&vrele_lock); 845 1.5.2.2 rmind gen = vrele_gen; 846 1.5.2.2 rmind while (vrele_pending && gen == vrele_gen) { 847 1.5.2.2 rmind cv_broadcast(&vrele_cv); 848 1.5.2.2 rmind cv_wait(&vrele_cv, &vrele_lock); 849 1.5.2.2 rmind } 850 1.5.2.2 rmind mutex_exit(&vrele_lock); 851 1.5.2.2 rmind } 852 1.5.2.2 rmind 853 1.5.2.2 rmind /* 854 1.5.2.2 rmind * Vnode reference, where a reference is already held by some other 855 1.5.2.2 rmind * object (for example, a file structure). 856 1.5.2.2 rmind */ 857 1.5.2.2 rmind void 858 1.5.2.2 rmind vref(vnode_t *vp) 859 1.5.2.2 rmind { 860 1.5.2.2 rmind 861 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 862 1.5.2.2 rmind KASSERT(vp->v_usecount != 0); 863 1.5.2.2 rmind 864 1.5.2.2 rmind atomic_inc_uint(&vp->v_usecount); 865 1.5.2.2 rmind } 866 1.5.2.2 rmind 867 1.5.2.2 rmind /* 868 1.5.2.2 rmind * Page or buffer structure gets a reference. 869 1.5.2.2 rmind * Called with v_interlock held. 870 1.5.2.2 rmind */ 871 1.5.2.2 rmind void 872 1.5.2.2 rmind vholdl(vnode_t *vp) 873 1.5.2.2 rmind { 874 1.5.2.2 rmind 875 1.5.2.2 rmind KASSERT(mutex_owned(vp->v_interlock)); 876 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 877 1.5.2.2 rmind 878 1.5.2.2 rmind if (vp->v_holdcnt++ == 0 && vp->v_usecount == 0) { 879 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 880 1.5.2.2 rmind KASSERT(vp->v_freelisthd == &vnode_free_list); 881 1.5.2.2 rmind TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 882 1.5.2.2 rmind vp->v_freelisthd = &vnode_hold_list; 883 1.5.2.2 rmind TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 884 1.5.2.2 rmind mutex_exit(&vnode_free_list_lock); 885 1.5.2.2 rmind } 886 1.5.2.2 rmind } 887 1.5.2.2 rmind 888 1.5.2.2 rmind /* 889 1.5.2.2 rmind * Page or buffer structure frees a reference. 890 1.5.2.2 rmind * Called with v_interlock held. 891 1.5.2.2 rmind */ 892 1.5.2.2 rmind void 893 1.5.2.2 rmind holdrelel(vnode_t *vp) 894 1.5.2.2 rmind { 895 1.5.2.2 rmind 896 1.5.2.2 rmind KASSERT(mutex_owned(vp->v_interlock)); 897 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 898 1.5.2.2 rmind 899 1.5.2.2 rmind if (vp->v_holdcnt <= 0) { 900 1.5.2.2 rmind vpanic(vp, "holdrelel: holdcnt vp %p"); 901 1.5.2.2 rmind } 902 1.5.2.2 rmind 903 1.5.2.2 rmind vp->v_holdcnt--; 904 1.5.2.2 rmind if (vp->v_holdcnt == 0 && vp->v_usecount == 0) { 905 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 906 1.5.2.2 rmind KASSERT(vp->v_freelisthd == &vnode_hold_list); 907 1.5.2.2 rmind TAILQ_REMOVE(vp->v_freelisthd, vp, v_freelist); 908 1.5.2.2 rmind vp->v_freelisthd = &vnode_free_list; 909 1.5.2.2 rmind TAILQ_INSERT_TAIL(vp->v_freelisthd, vp, v_freelist); 910 1.5.2.2 rmind mutex_exit(&vnode_free_list_lock); 911 1.5.2.2 rmind } 912 1.5.2.2 rmind } 913 1.5.2.2 rmind 914 1.5.2.2 rmind /* 915 1.5.2.2 rmind * Disassociate the underlying file system from a vnode. 916 1.5.2.2 rmind * 917 1.5.2.2 rmind * Must be called with the interlock held, and will return with it held. 918 1.5.2.2 rmind */ 919 1.5.2.2 rmind void 920 1.5.2.2 rmind vclean(vnode_t *vp, int flags) 921 1.5.2.2 rmind { 922 1.5.2.2 rmind lwp_t *l = curlwp; 923 1.5.2.2 rmind bool recycle, active; 924 1.5.2.2 rmind int error; 925 1.5.2.2 rmind 926 1.5.2.2 rmind KASSERT(mutex_owned(vp->v_interlock)); 927 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 928 1.5.2.2 rmind KASSERT(vp->v_usecount != 0); 929 1.5.2.2 rmind 930 1.5.2.2 rmind /* If cleaning is already in progress wait until done and return. */ 931 1.5.2.2 rmind if (vp->v_iflag & VI_XLOCK) { 932 1.5.2.2 rmind vwait(vp, VI_XLOCK); 933 1.5.2.2 rmind return; 934 1.5.2.2 rmind } 935 1.5.2.2 rmind 936 1.5.2.2 rmind /* If already clean, nothing to do. */ 937 1.5.2.2 rmind if ((vp->v_iflag & VI_CLEAN) != 0) { 938 1.5.2.2 rmind return; 939 1.5.2.2 rmind } 940 1.5.2.2 rmind 941 1.5.2.2 rmind /* 942 1.5.2.2 rmind * Prevent the vnode from being recycled or brought into use 943 1.5.2.2 rmind * while we clean it out. 944 1.5.2.2 rmind */ 945 1.5.2.2 rmind vp->v_iflag |= VI_XLOCK; 946 1.5.2.2 rmind if (vp->v_iflag & VI_EXECMAP) { 947 1.5.2.2 rmind atomic_add_int(&uvmexp.execpages, -vp->v_uobj.uo_npages); 948 1.5.2.2 rmind atomic_add_int(&uvmexp.filepages, vp->v_uobj.uo_npages); 949 1.5.2.2 rmind } 950 1.5.2.2 rmind vp->v_iflag &= ~(VI_TEXT|VI_EXECMAP); 951 1.5.2.2 rmind active = (vp->v_usecount & VC_MASK) > 1; 952 1.5.2.2 rmind 953 1.5.2.2 rmind /* XXXAD should not lock vnode under layer */ 954 1.5.2.2 rmind mutex_exit(vp->v_interlock); 955 1.5.2.2 rmind VOP_LOCK(vp, LK_EXCLUSIVE); 956 1.5.2.2 rmind 957 1.5.2.2 rmind /* 958 1.5.2.2 rmind * Clean out any cached data associated with the vnode. 959 1.5.2.2 rmind * If purging an active vnode, it must be closed and 960 1.5.2.2 rmind * deactivated before being reclaimed. Note that the 961 1.5.2.2 rmind * VOP_INACTIVE will unlock the vnode. 962 1.5.2.2 rmind */ 963 1.5.2.2 rmind if (flags & DOCLOSE) { 964 1.5.2.2 rmind error = vinvalbuf(vp, V_SAVE, NOCRED, l, 0, 0); 965 1.5.2.2 rmind if (error != 0) { 966 1.5.2.2 rmind /* XXX, fix vn_start_write's grab of mp and use that. */ 967 1.5.2.2 rmind 968 1.5.2.2 rmind if (wapbl_vphaswapbl(vp)) 969 1.5.2.2 rmind WAPBL_DISCARD(wapbl_vptomp(vp)); 970 1.5.2.2 rmind error = vinvalbuf(vp, 0, NOCRED, l, 0, 0); 971 1.5.2.2 rmind } 972 1.5.2.2 rmind KASSERT(error == 0); 973 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_ONWORKLST) == 0); 974 1.5.2.2 rmind if (active && (vp->v_type == VBLK || vp->v_type == VCHR)) { 975 1.5.2.2 rmind spec_node_revoke(vp); 976 1.5.2.2 rmind } 977 1.5.2.2 rmind } 978 1.5.2.2 rmind if (active) { 979 1.5.2.2 rmind VOP_INACTIVE(vp, &recycle); 980 1.5.2.2 rmind } else { 981 1.5.2.2 rmind /* 982 1.5.2.2 rmind * Any other processes trying to obtain this lock must first 983 1.5.2.2 rmind * wait for VI_XLOCK to clear, then call the new lock operation. 984 1.5.2.2 rmind */ 985 1.5.2.2 rmind VOP_UNLOCK(vp); 986 1.5.2.2 rmind } 987 1.5.2.2 rmind 988 1.5.2.2 rmind /* Disassociate the underlying file system from the vnode. */ 989 1.5.2.2 rmind if (VOP_RECLAIM(vp)) { 990 1.5.2.2 rmind vpanic(vp, "vclean: cannot reclaim"); 991 1.5.2.2 rmind } 992 1.5.2.2 rmind 993 1.5.2.2 rmind KASSERT(vp->v_uobj.uo_npages == 0); 994 1.5.2.2 rmind if (vp->v_type == VREG && vp->v_ractx != NULL) { 995 1.5.2.2 rmind uvm_ra_freectx(vp->v_ractx); 996 1.5.2.2 rmind vp->v_ractx = NULL; 997 1.5.2.2 rmind } 998 1.5.2.2 rmind cache_purge(vp); 999 1.5.2.2 rmind 1000 1.5.2.2 rmind /* Done with purge, notify sleepers of the grim news. */ 1001 1.5.2.2 rmind mutex_enter(vp->v_interlock); 1002 1.5.2.2 rmind vp->v_op = dead_vnodeop_p; 1003 1.5.2.2 rmind vp->v_tag = VT_NON; 1004 1.5.2.2 rmind KNOTE(&vp->v_klist, NOTE_REVOKE); 1005 1.5.2.2 rmind vp->v_iflag &= ~VI_XLOCK; 1006 1.5.2.2 rmind vp->v_vflag &= ~VV_LOCKSWORK; 1007 1.5.2.2 rmind if ((flags & DOCLOSE) != 0) { 1008 1.5.2.2 rmind vp->v_iflag |= VI_CLEAN; 1009 1.5.2.2 rmind } 1010 1.5.2.2 rmind cv_broadcast(&vp->v_cv); 1011 1.5.2.2 rmind 1012 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_ONWORKLST) == 0); 1013 1.5.2.2 rmind } 1014 1.5.2.2 rmind 1015 1.5.2.2 rmind /* 1016 1.5.2.2 rmind * Recycle an unused vnode to the front of the free list. 1017 1.5.2.2 rmind * Release the passed interlock if the vnode will be recycled. 1018 1.5.2.2 rmind */ 1019 1.5.2.2 rmind int 1020 1.5.2.2 rmind vrecycle(vnode_t *vp, kmutex_t *inter_lkp, struct lwp *l) 1021 1.5.2.2 rmind { 1022 1.5.2.2 rmind 1023 1.5.2.2 rmind KASSERT((vp->v_iflag & VI_MARKER) == 0); 1024 1.5.2.2 rmind 1025 1.5.2.2 rmind mutex_enter(vp->v_interlock); 1026 1.5.2.2 rmind if (vp->v_usecount != 0) { 1027 1.5.2.2 rmind mutex_exit(vp->v_interlock); 1028 1.5.2.2 rmind return 0; 1029 1.5.2.2 rmind } 1030 1.5.2.2 rmind if (inter_lkp) { 1031 1.5.2.2 rmind mutex_exit(inter_lkp); 1032 1.5.2.2 rmind } 1033 1.5.2.2 rmind vremfree(vp); 1034 1.5.2.2 rmind vp->v_usecount = 1; 1035 1.5.2.2 rmind vclean(vp, DOCLOSE); 1036 1.5.2.2 rmind vrelel(vp, 0); 1037 1.5.2.2 rmind return 1; 1038 1.5.2.2 rmind } 1039 1.5.2.2 rmind 1040 1.5.2.2 rmind /* 1041 1.5.2.2 rmind * Eliminate all activity associated with the requested vnode 1042 1.5.2.2 rmind * and with all vnodes aliased to the requested vnode. 1043 1.5.2.2 rmind */ 1044 1.5.2.2 rmind void 1045 1.5.2.2 rmind vrevoke(vnode_t *vp) 1046 1.5.2.2 rmind { 1047 1.5.2.2 rmind vnode_t *vq, **vpp; 1048 1.5.2.2 rmind enum vtype type; 1049 1.5.2.2 rmind dev_t dev; 1050 1.5.2.2 rmind 1051 1.5.2.2 rmind KASSERT(vp->v_usecount > 0); 1052 1.5.2.2 rmind 1053 1.5.2.2 rmind mutex_enter(vp->v_interlock); 1054 1.5.2.2 rmind if ((vp->v_iflag & VI_CLEAN) != 0) { 1055 1.5.2.2 rmind mutex_exit(vp->v_interlock); 1056 1.5.2.2 rmind return; 1057 1.5.2.2 rmind } else if (vp->v_type != VBLK && vp->v_type != VCHR) { 1058 1.5.2.2 rmind atomic_inc_uint(&vp->v_usecount); 1059 1.5.2.2 rmind vclean(vp, DOCLOSE); 1060 1.5.2.2 rmind vrelel(vp, 0); 1061 1.5.2.2 rmind return; 1062 1.5.2.2 rmind } else { 1063 1.5.2.2 rmind dev = vp->v_rdev; 1064 1.5.2.2 rmind type = vp->v_type; 1065 1.5.2.2 rmind mutex_exit(vp->v_interlock); 1066 1.5.2.2 rmind } 1067 1.5.2.2 rmind 1068 1.5.2.2 rmind vpp = &specfs_hash[SPECHASH(dev)]; 1069 1.5.2.2 rmind mutex_enter(&device_lock); 1070 1.5.2.2 rmind for (vq = *vpp; vq != NULL;) { 1071 1.5.2.2 rmind /* If clean or being cleaned, then ignore it. */ 1072 1.5.2.2 rmind mutex_enter(vq->v_interlock); 1073 1.5.2.2 rmind if ((vq->v_iflag & (VI_CLEAN | VI_XLOCK)) != 0 || 1074 1.5.2.2 rmind vq->v_rdev != dev || vq->v_type != type) { 1075 1.5.2.2 rmind mutex_exit(vq->v_interlock); 1076 1.5.2.2 rmind vq = vq->v_specnext; 1077 1.5.2.2 rmind continue; 1078 1.5.2.2 rmind } 1079 1.5.2.2 rmind mutex_exit(&device_lock); 1080 1.5.2.2 rmind if (vq->v_usecount == 0) { 1081 1.5.2.2 rmind vremfree(vq); 1082 1.5.2.2 rmind vq->v_usecount = 1; 1083 1.5.2.2 rmind } else { 1084 1.5.2.2 rmind atomic_inc_uint(&vq->v_usecount); 1085 1.5.2.2 rmind } 1086 1.5.2.2 rmind vclean(vq, DOCLOSE); 1087 1.5.2.2 rmind vrelel(vq, 0); 1088 1.5.2.2 rmind mutex_enter(&device_lock); 1089 1.5.2.2 rmind vq = *vpp; 1090 1.5.2.2 rmind } 1091 1.5.2.2 rmind mutex_exit(&device_lock); 1092 1.5.2.2 rmind } 1093 1.5.2.2 rmind 1094 1.5.2.2 rmind /* 1095 1.5.2.2 rmind * Eliminate all activity associated with a vnode in preparation for 1096 1.5.2.2 rmind * reuse. Drops a reference from the vnode. 1097 1.5.2.2 rmind */ 1098 1.5.2.2 rmind void 1099 1.5.2.2 rmind vgone(vnode_t *vp) 1100 1.5.2.2 rmind { 1101 1.5.2.2 rmind 1102 1.5.2.2 rmind mutex_enter(vp->v_interlock); 1103 1.5.2.2 rmind vclean(vp, DOCLOSE); 1104 1.5.2.2 rmind vrelel(vp, 0); 1105 1.5.2.2 rmind } 1106 1.5.2.2 rmind 1107 1.5.2.2 rmind /* 1108 1.5.2.2 rmind * Update outstanding I/O count and do wakeup if requested. 1109 1.5.2.2 rmind */ 1110 1.5.2.2 rmind void 1111 1.5.2.2 rmind vwakeup(struct buf *bp) 1112 1.5.2.2 rmind { 1113 1.5.2.2 rmind vnode_t *vp; 1114 1.5.2.2 rmind 1115 1.5.2.2 rmind if ((vp = bp->b_vp) == NULL) 1116 1.5.2.2 rmind return; 1117 1.5.2.2 rmind 1118 1.5.2.2 rmind KASSERT(bp->b_objlock == vp->v_interlock); 1119 1.5.2.2 rmind KASSERT(mutex_owned(bp->b_objlock)); 1120 1.5.2.2 rmind 1121 1.5.2.2 rmind if (--vp->v_numoutput < 0) 1122 1.5.2.2 rmind panic("vwakeup: neg numoutput, vp %p", vp); 1123 1.5.2.2 rmind if (vp->v_numoutput == 0) 1124 1.5.2.2 rmind cv_broadcast(&vp->v_cv); 1125 1.5.2.2 rmind } 1126 1.5.2.2 rmind 1127 1.5.2.2 rmind /* 1128 1.5.2.2 rmind * Wait for a vnode (typically with VI_XLOCK set) to be cleaned or 1129 1.5.2.2 rmind * recycled. 1130 1.5.2.2 rmind */ 1131 1.5.2.2 rmind void 1132 1.5.2.2 rmind vwait(vnode_t *vp, int flags) 1133 1.5.2.2 rmind { 1134 1.5.2.2 rmind 1135 1.5.2.2 rmind KASSERT(mutex_owned(vp->v_interlock)); 1136 1.5.2.2 rmind KASSERT(vp->v_usecount != 0); 1137 1.5.2.2 rmind 1138 1.5.2.2 rmind while ((vp->v_iflag & flags) != 0) 1139 1.5.2.2 rmind cv_wait(&vp->v_cv, vp->v_interlock); 1140 1.5.2.2 rmind } 1141 1.5.2.2 rmind 1142 1.5.2.2 rmind int 1143 1.5.2.2 rmind vfs_drainvnodes(long target) 1144 1.5.2.2 rmind { 1145 1.5.2.2 rmind 1146 1.5.2.2 rmind while (numvnodes > target) { 1147 1.5.2.2 rmind vnode_t *vp; 1148 1.5.2.2 rmind 1149 1.5.2.2 rmind mutex_enter(&vnode_free_list_lock); 1150 1.5.2.2 rmind vp = getcleanvnode(); 1151 1.5.2.2 rmind if (vp == NULL) { 1152 1.5.2.2 rmind return EBUSY; 1153 1.5.2.2 rmind } 1154 1.5.2.2 rmind ungetnewvnode(vp); 1155 1.5.2.2 rmind } 1156 1.5.2.2 rmind return 0; 1157 1.5.2.2 rmind } 1158 1.5.2.2 rmind 1159 1.5.2.2 rmind void 1160 1.5.2.2 rmind vpanic(vnode_t *vp, const char *msg) 1161 1.5.2.2 rmind { 1162 1.5.2.2 rmind #ifdef DIAGNOSTIC 1163 1.5.2.2 rmind 1164 1.5.2.2 rmind vprint(NULL, vp); 1165 1.5.2.2 rmind panic("%s\n", msg); 1166 1.5.2.2 rmind #endif 1167 1.5.2.2 rmind } 1168
Indexes created Sat Sep 20 22:09:52 GMT 2025