layer_subr.c revision 1.25 1 1.25 ad /* $NetBSD: layer_subr.c,v 1.25 2008/01/24 17:32:55 ad Exp $ */
2 1.1 wrstuden
3 1.1 wrstuden /*
4 1.1 wrstuden * Copyright (c) 1999 National Aeronautics & Space Administration
5 1.1 wrstuden * All rights reserved.
6 1.1 wrstuden *
7 1.1 wrstuden * This software was written by William Studenmund of the
8 1.8 wiz * Numerical Aerospace Simulation Facility, NASA Ames Research Center.
9 1.1 wrstuden *
10 1.1 wrstuden * Redistribution and use in source and binary forms, with or without
11 1.1 wrstuden * modification, are permitted provided that the following conditions
12 1.1 wrstuden * are met:
13 1.1 wrstuden * 1. Redistributions of source code must retain the above copyright
14 1.1 wrstuden * notice, this list of conditions and the following disclaimer.
15 1.1 wrstuden * 2. Redistributions in binary form must reproduce the above copyright
16 1.1 wrstuden * notice, this list of conditions and the following disclaimer in the
17 1.1 wrstuden * documentation and/or other materials provided with the distribution.
18 1.5 soren * 3. Neither the name of the National Aeronautics & Space Administration
19 1.1 wrstuden * nor the names of its contributors may be used to endorse or promote
20 1.1 wrstuden * products derived from this software without specific prior written
21 1.1 wrstuden * permission.
22 1.1 wrstuden *
23 1.1 wrstuden * THIS SOFTWARE IS PROVIDED BY THE NATIONAL AERONAUTICS & SPACE ADMINISTRATION
24 1.1 wrstuden * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
25 1.1 wrstuden * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
26 1.1 wrstuden * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE ADMINISTRATION OR CONTRIB-
27 1.1 wrstuden * UTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
28 1.1 wrstuden * OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 1.1 wrstuden * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
30 1.1 wrstuden * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
31 1.1 wrstuden * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
32 1.1 wrstuden * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33 1.1 wrstuden * POSSIBILITY OF SUCH DAMAGE.
34 1.1 wrstuden */
35 1.1 wrstuden /*
36 1.1 wrstuden * Copyright (c) 1992, 1993
37 1.1 wrstuden * The Regents of the University of California. All rights reserved.
38 1.1 wrstuden *
39 1.1 wrstuden * This code is derived from software donated to Berkeley by
40 1.1 wrstuden * Jan-Simon Pendry.
41 1.1 wrstuden *
42 1.1 wrstuden * Redistribution and use in source and binary forms, with or without
43 1.1 wrstuden * modification, are permitted provided that the following conditions
44 1.1 wrstuden * are met:
45 1.1 wrstuden * 1. Redistributions of source code must retain the above copyright
46 1.1 wrstuden * notice, this list of conditions and the following disclaimer.
47 1.1 wrstuden * 2. Redistributions in binary form must reproduce the above copyright
48 1.1 wrstuden * notice, this list of conditions and the following disclaimer in the
49 1.1 wrstuden * documentation and/or other materials provided with the distribution.
50 1.15 agc * 3. Neither the name of the University nor the names of its contributors
51 1.1 wrstuden * may be used to endorse or promote products derived from this software
52 1.1 wrstuden * without specific prior written permission.
53 1.1 wrstuden *
54 1.1 wrstuden * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 1.1 wrstuden * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 1.1 wrstuden * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 1.1 wrstuden * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 1.1 wrstuden * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 1.1 wrstuden * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 1.1 wrstuden * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 1.1 wrstuden * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 1.1 wrstuden * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 1.1 wrstuden * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 1.1 wrstuden * SUCH DAMAGE.
65 1.1 wrstuden *
66 1.1 wrstuden * from: Id: lofs_subr.c,v 1.11 1992/05/30 10:05:43 jsp Exp
67 1.1 wrstuden * @(#)null_subr.c 8.7 (Berkeley) 5/14/95
68 1.1 wrstuden */
69 1.9 lukem
70 1.9 lukem #include <sys/cdefs.h>
71 1.25 ad __KERNEL_RCSID(0, "$NetBSD: layer_subr.c,v 1.25 2008/01/24 17:32:55 ad Exp $");
72 1.1 wrstuden
73 1.1 wrstuden #include <sys/param.h>
74 1.1 wrstuden #include <sys/systm.h>
75 1.1 wrstuden #include <sys/proc.h>
76 1.1 wrstuden #include <sys/time.h>
77 1.1 wrstuden #include <sys/vnode.h>
78 1.1 wrstuden #include <sys/mount.h>
79 1.1 wrstuden #include <sys/namei.h>
80 1.23 ad #include <sys/kmem.h>
81 1.1 wrstuden #include <sys/malloc.h>
82 1.23 ad
83 1.1 wrstuden #include <miscfs/specfs/specdev.h>
84 1.1 wrstuden #include <miscfs/genfs/layer.h>
85 1.1 wrstuden #include <miscfs/genfs/layer_extern.h>
86 1.1 wrstuden
87 1.1 wrstuden #define NLAYERNODECACHE 16
88 1.1 wrstuden
89 1.16 erh #ifdef LAYERFS_DIAGNOSTIC
90 1.16 erh int layerfs_debug = 1;
91 1.16 erh #endif
92 1.16 erh
93 1.1 wrstuden /*
94 1.1 wrstuden * layer cache:
95 1.1 wrstuden * Each cache entry holds a reference to the lower vnode
96 1.1 wrstuden * along with a pointer to the alias vnode. When an
97 1.1 wrstuden * entry is added the lower vnode is VREF'd. When the
98 1.1 wrstuden * alias is removed the lower vnode is vrele'd.
99 1.1 wrstuden */
100 1.1 wrstuden
101 1.1 wrstuden /*
102 1.1 wrstuden * Initialise cache headers
103 1.1 wrstuden */
104 1.1 wrstuden void
105 1.1 wrstuden layerfs_init()
106 1.1 wrstuden {
107 1.6 jdolecek #ifdef LAYERFS_DIAGNOSTIC
108 1.16 erh if (layerfs_debug)
109 1.16 erh printf("layerfs_init\n"); /* printed during system boot */
110 1.6 jdolecek #endif
111 1.6 jdolecek }
112 1.1 wrstuden
113 1.6 jdolecek /*
114 1.6 jdolecek * Free global resources of layerfs.
115 1.6 jdolecek */
116 1.6 jdolecek void
117 1.6 jdolecek layerfs_done()
118 1.6 jdolecek {
119 1.1 wrstuden #ifdef LAYERFS_DIAGNOSTIC
120 1.16 erh if (layerfs_debug)
121 1.16 erh printf("layerfs_done\n"); /* printed on layerfs detach */
122 1.1 wrstuden #endif
123 1.1 wrstuden }
124 1.1 wrstuden
125 1.1 wrstuden /*
126 1.21 chs * Return a locked, VREF'ed alias for lower vnode if already exists, else NULL.
127 1.21 chs * The layermp's hashlock must be held on entry.
128 1.21 chs * It will be held upon return iff we return NULL.
129 1.1 wrstuden */
130 1.1 wrstuden struct vnode *
131 1.1 wrstuden layer_node_find(mp, lowervp)
132 1.1 wrstuden struct mount *mp;
133 1.1 wrstuden struct vnode *lowervp;
134 1.1 wrstuden {
135 1.1 wrstuden struct layer_mount *lmp = MOUNTTOLAYERMOUNT(mp);
136 1.1 wrstuden struct layer_node_hashhead *hd;
137 1.1 wrstuden struct layer_node *a;
138 1.1 wrstuden struct vnode *vp;
139 1.21 chs int error;
140 1.1 wrstuden
141 1.1 wrstuden /*
142 1.1 wrstuden * Find hash base, and then search the (two-way) linked
143 1.1 wrstuden * list looking for a layer_node structure which is referencing
144 1.1 wrstuden * the lower vnode. If found, the increment the layer_node
145 1.1 wrstuden * reference count (but NOT the lower vnode's VREF counter)
146 1.1 wrstuden * and return the vnode locked.
147 1.1 wrstuden */
148 1.1 wrstuden hd = LAYER_NHASH(lmp, lowervp);
149 1.1 wrstuden loop:
150 1.21 chs LIST_FOREACH(a, hd, layer_hash) {
151 1.1 wrstuden if (a->layer_lowervp == lowervp && LAYERTOV(a)->v_mount == mp) {
152 1.1 wrstuden vp = LAYERTOV(a);
153 1.24 ad mutex_enter(&vp->v_interlock);
154 1.1 wrstuden /*
155 1.24 ad * If we find a node being cleaned out, then
156 1.24 ad * ignore it and continue. A thread trying to
157 1.24 ad * clean out the extant layer vnode needs to
158 1.24 ad * acquire the shared lock (i.e. the lower
159 1.24 ad * vnode's lock), which our caller already holds.
160 1.24 ad * To allow the cleaning to succeed the current
161 1.24 ad * thread must make progress. So, for a brief
162 1.24 ad * time more than one vnode in a layered file
163 1.24 ad * system may refer to a single vnode in the
164 1.24 ad * lower file system.
165 1.1 wrstuden */
166 1.24 ad if ((vp->v_iflag & VI_XLOCK) != 0) {
167 1.24 ad mutex_exit(&vp->v_interlock);
168 1.24 ad continue;
169 1.24 ad }
170 1.22 ad mutex_exit(&lmp->layerm_hashlock);
171 1.24 ad /*
172 1.24 ad * We must not let vget() try to lock the layer
173 1.24 ad * vp, since the lower vp is already locked and
174 1.24 ad * locking the layer vp will involve locking
175 1.24 ad * the lower vp (whether or not they actually
176 1.24 ad * share a lock). Instead, take the layer vp's
177 1.24 ad * lock separately afterward, but only if it
178 1.24 ad * does not share the lower vp's lock.
179 1.24 ad */
180 1.24 ad error = vget(vp, LK_INTERLOCK | LK_NOWAIT);
181 1.21 chs if (error) {
182 1.24 ad kpause("layerfs", false, 1, NULL);
183 1.22 ad mutex_enter(&lmp->layerm_hashlock);
184 1.1 wrstuden goto loop;
185 1.21 chs }
186 1.21 chs LAYERFS_UPPERLOCK(vp, LK_EXCLUSIVE, error);
187 1.1 wrstuden return (vp);
188 1.1 wrstuden }
189 1.1 wrstuden }
190 1.1 wrstuden return NULL;
191 1.1 wrstuden }
192 1.1 wrstuden
193 1.1 wrstuden
194 1.1 wrstuden /*
195 1.1 wrstuden * Make a new layer_node node.
196 1.1 wrstuden * Vp is the alias vnode, lowervp is the lower vnode.
197 1.1 wrstuden * Maintain a reference to lowervp.
198 1.1 wrstuden */
199 1.1 wrstuden int
200 1.1 wrstuden layer_node_alloc(mp, lowervp, vpp)
201 1.1 wrstuden struct mount *mp;
202 1.1 wrstuden struct vnode *lowervp;
203 1.1 wrstuden struct vnode **vpp;
204 1.1 wrstuden {
205 1.1 wrstuden struct layer_mount *lmp = MOUNTTOLAYERMOUNT(mp);
206 1.1 wrstuden struct layer_node_hashhead *hd;
207 1.1 wrstuden struct layer_node *xp;
208 1.1 wrstuden struct vnode *vp, *nvp;
209 1.1 wrstuden int error;
210 1.17 xtraeme extern int (**dead_vnodeop_p)(void *);
211 1.1 wrstuden
212 1.23 ad error = getnewvnode(lmp->layerm_tag, mp, lmp->layerm_vnodeop_p, &vp);
213 1.23 ad if (error != 0)
214 1.1 wrstuden return (error);
215 1.1 wrstuden vp->v_type = lowervp->v_type;
216 1.23 ad mutex_enter(&vp->v_interlock);
217 1.22 ad vp->v_iflag |= VI_LAYER;
218 1.23 ad mutex_exit(&vp->v_interlock);
219 1.1 wrstuden
220 1.23 ad xp = kmem_alloc(lmp->layerm_size, KM_SLEEP);
221 1.23 ad if (xp == NULL) {
222 1.23 ad ungetnewvnode(vp);
223 1.23 ad return ENOMEM;
224 1.23 ad }
225 1.1 wrstuden if (vp->v_type == VBLK || vp->v_type == VCHR) {
226 1.25 ad spec_node_init(vp, lowervp->v_rdev);
227 1.1 wrstuden }
228 1.1 wrstuden
229 1.1 wrstuden vp->v_data = xp;
230 1.23 ad vp->v_vflag = (vp->v_vflag & ~VV_MPSAFE) |
231 1.23 ad (lowervp->v_vflag & VV_MPSAFE);
232 1.1 wrstuden xp->layer_vnode = vp;
233 1.1 wrstuden xp->layer_lowervp = lowervp;
234 1.1 wrstuden xp->layer_flags = 0;
235 1.21 chs
236 1.1 wrstuden /*
237 1.1 wrstuden * Before we insert our new node onto the hash chains,
238 1.1 wrstuden * check to see if someone else has beaten us to it.
239 1.1 wrstuden * (We could have slept in MALLOC.)
240 1.1 wrstuden */
241 1.22 ad mutex_enter(&lmp->layerm_hashlock);
242 1.1 wrstuden if ((nvp = layer_node_find(mp, lowervp)) != NULL) {
243 1.1 wrstuden *vpp = nvp;
244 1.1 wrstuden
245 1.1 wrstuden /* free the substructures we've allocated. */
246 1.23 ad kmem_free(xp, lmp->layerm_size);
247 1.1 wrstuden if (vp->v_type == VBLK || vp->v_type == VCHR)
248 1.25 ad spec_node_destroy(vp);
249 1.1 wrstuden
250 1.1 wrstuden vp->v_type = VBAD; /* node is discarded */
251 1.1 wrstuden vp->v_op = dead_vnodeop_p; /* so ops will still work */
252 1.1 wrstuden vrele(vp); /* get rid of it. */
253 1.1 wrstuden return (0);
254 1.1 wrstuden }
255 1.1 wrstuden
256 1.1 wrstuden /*
257 1.1 wrstuden * Now lock the new node. We rely on the fact that we were passed
258 1.1 wrstuden * a locked vnode. If the lower node is exporting a struct lock
259 1.1 wrstuden * (v_vnlock != NULL) then we just set the upper v_vnlock to the
260 1.1 wrstuden * lower one, and both are now locked. If the lower node is exporting
261 1.1 wrstuden * NULL, then we copy that up and manually lock the upper node.
262 1.1 wrstuden *
263 1.1 wrstuden * LAYERFS_UPPERLOCK already has the test, so we use it after copying
264 1.1 wrstuden * up the v_vnlock from below.
265 1.1 wrstuden */
266 1.1 wrstuden
267 1.1 wrstuden vp->v_vnlock = lowervp->v_vnlock;
268 1.1 wrstuden LAYERFS_UPPERLOCK(vp, LK_EXCLUSIVE, error);
269 1.21 chs KASSERT(error == 0);
270 1.1 wrstuden
271 1.1 wrstuden /*
272 1.21 chs * Insert the new node into the hash.
273 1.21 chs * Add a reference to the lower node.
274 1.1 wrstuden */
275 1.1 wrstuden
276 1.1 wrstuden *vpp = vp;
277 1.21 chs VREF(lowervp);
278 1.1 wrstuden hd = LAYER_NHASH(lmp, lowervp);
279 1.1 wrstuden LIST_INSERT_HEAD(hd, xp, layer_hash);
280 1.7 chs uvm_vnp_setsize(vp, 0);
281 1.22 ad mutex_exit(&lmp->layerm_hashlock);
282 1.1 wrstuden return (0);
283 1.1 wrstuden }
284 1.1 wrstuden
285 1.1 wrstuden
286 1.1 wrstuden /*
287 1.1 wrstuden * Try to find an existing layer_node vnode refering
288 1.1 wrstuden * to it, otherwise make a new layer_node vnode which
289 1.1 wrstuden * contains a reference to the lower vnode.
290 1.1 wrstuden *
291 1.1 wrstuden * >>> we assume that the lower node is already locked upon entry, so we
292 1.1 wrstuden * propagate the lock state to upper node <<
293 1.1 wrstuden */
294 1.1 wrstuden int
295 1.1 wrstuden layer_node_create(mp, lowervp, newvpp)
296 1.1 wrstuden struct mount *mp;
297 1.1 wrstuden struct vnode *lowervp;
298 1.1 wrstuden struct vnode **newvpp;
299 1.1 wrstuden {
300 1.1 wrstuden struct vnode *aliasvp;
301 1.1 wrstuden struct layer_mount *lmp = MOUNTTOLAYERMOUNT(mp);
302 1.1 wrstuden
303 1.22 ad mutex_enter(&lmp->layerm_hashlock);
304 1.21 chs aliasvp = layer_node_find(mp, lowervp);
305 1.21 chs if (aliasvp != NULL) {
306 1.1 wrstuden /*
307 1.1 wrstuden * layer_node_find has taken another reference
308 1.1 wrstuden * to the alias vnode and moved the lock holding to
309 1.1 wrstuden * aliasvp
310 1.1 wrstuden */
311 1.1 wrstuden #ifdef LAYERFS_DIAGNOSTIC
312 1.16 erh if (layerfs_debug)
313 1.16 erh vprint("layer_node_create: exists", aliasvp);
314 1.1 wrstuden #endif
315 1.1 wrstuden } else {
316 1.1 wrstuden int error;
317 1.1 wrstuden
318 1.22 ad mutex_exit(&lmp->layerm_hashlock);
319 1.21 chs
320 1.1 wrstuden /*
321 1.1 wrstuden * Get new vnode.
322 1.1 wrstuden */
323 1.1 wrstuden #ifdef LAYERFS_DIAGNOSTIC
324 1.16 erh if (layerfs_debug)
325 1.16 erh printf("layer_node_create: create new alias vnode\n");
326 1.1 wrstuden #endif
327 1.1 wrstuden
328 1.1 wrstuden /*
329 1.1 wrstuden * Make new vnode reference the layer_node.
330 1.1 wrstuden */
331 1.1 wrstuden if ((error = (lmp->layerm_alloc)(mp, lowervp, &aliasvp)) != 0)
332 1.1 wrstuden return error;
333 1.1 wrstuden
334 1.1 wrstuden /*
335 1.1 wrstuden * aliasvp is already VREF'd by getnewvnode()
336 1.1 wrstuden */
337 1.1 wrstuden }
338 1.1 wrstuden
339 1.1 wrstuden /*
340 1.1 wrstuden * Now that we have VREF'd the upper vnode, release the reference
341 1.19 wiz * to the lower node. The existence of the layer_node retains one
342 1.1 wrstuden * reference to the lower node.
343 1.1 wrstuden */
344 1.1 wrstuden vrele(lowervp);
345 1.1 wrstuden
346 1.1 wrstuden #ifdef DIAGNOSTIC
347 1.1 wrstuden if (lowervp->v_usecount < 1) {
348 1.1 wrstuden /* Should never happen... */
349 1.1 wrstuden vprint("layer_node_create: alias", aliasvp);
350 1.1 wrstuden vprint("layer_node_create: lower", lowervp);
351 1.1 wrstuden panic("layer_node_create: lower has 0 usecount.");
352 1.21 chs }
353 1.1 wrstuden #endif
354 1.1 wrstuden
355 1.1 wrstuden #ifdef LAYERFS_DIAGNOSTIC
356 1.16 erh if (layerfs_debug)
357 1.16 erh vprint("layer_node_create: alias", aliasvp);
358 1.1 wrstuden #endif
359 1.1 wrstuden *newvpp = aliasvp;
360 1.1 wrstuden return (0);
361 1.1 wrstuden }
362 1.1 wrstuden
363 1.16 erh #ifdef LAYERFS_DIAGNOSTIC
364 1.1 wrstuden struct vnode *
365 1.1 wrstuden layer_checkvp(vp, fil, lno)
366 1.1 wrstuden struct vnode *vp;
367 1.16 erh const char *fil;
368 1.1 wrstuden int lno;
369 1.1 wrstuden {
370 1.1 wrstuden struct layer_node *a = VTOLAYER(vp);
371 1.1 wrstuden #ifdef notyet
372 1.1 wrstuden /*
373 1.1 wrstuden * Can't do this check because vop_reclaim runs
374 1.1 wrstuden * with a funny vop vector.
375 1.1 wrstuden *
376 1.1 wrstuden * WRS - no it doesnt...
377 1.1 wrstuden */
378 1.1 wrstuden if (vp->v_op != layer_vnodeop_p) {
379 1.1 wrstuden printf ("layer_checkvp: on non-layer-node\n");
380 1.1 wrstuden #ifdef notyet
381 1.1 wrstuden while (layer_checkvp_barrier) /*WAIT*/ ;
382 1.1 wrstuden #endif
383 1.1 wrstuden panic("layer_checkvp");
384 1.1 wrstuden };
385 1.1 wrstuden #endif
386 1.1 wrstuden if (a->layer_lowervp == NULL) {
387 1.1 wrstuden /* Should never happen */
388 1.1 wrstuden int i; u_long *p;
389 1.1 wrstuden printf("vp = %p, ZERO ptr\n", vp);
390 1.1 wrstuden for (p = (u_long *) a, i = 0; i < 8; i++)
391 1.1 wrstuden printf(" %lx", p[i]);
392 1.1 wrstuden printf("\n");
393 1.1 wrstuden /* wait for debugger */
394 1.1 wrstuden panic("layer_checkvp");
395 1.1 wrstuden }
396 1.1 wrstuden if (a->layer_lowervp->v_usecount < 1) {
397 1.1 wrstuden int i; u_long *p;
398 1.1 wrstuden printf("vp = %p, unref'ed lowervp\n", vp);
399 1.1 wrstuden for (p = (u_long *) a, i = 0; i < 8; i++)
400 1.1 wrstuden printf(" %lx", p[i]);
401 1.1 wrstuden printf("\n");
402 1.1 wrstuden /* wait for debugger */
403 1.1 wrstuden panic ("layer with unref'ed lowervp");
404 1.1 wrstuden };
405 1.1 wrstuden #ifdef notnow
406 1.1 wrstuden printf("layer %p/%d -> %p/%d [%s, %d]\n",
407 1.1 wrstuden LAYERTOV(a), LAYERTOV(a)->v_usecount,
408 1.1 wrstuden a->layer_lowervp, a->layer_lowervp->v_usecount,
409 1.1 wrstuden fil, lno);
410 1.1 wrstuden #endif
411 1.1 wrstuden return a->layer_lowervp;
412 1.1 wrstuden }
413 1.16 erh #endif
414