xref: /freebsd/sys/geom/geom_vfs.c (revision 8ddb146abcdf061be9f2c0db7e391697dafad85c)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2004 Poul-Henning Kamp
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/bio.h>
35 #include <sys/kernel.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/mutex.h>
39 #include <sys/sbuf.h>
40 #include <sys/vnode.h>
41 #include <sys/mount.h>
42 
43 #include <geom/geom.h>
44 #include <geom/geom_vfs.h>
45 
46 /*
47  * subroutines for use by filesystems.
48  *
49  * XXX: should maybe live somewhere else ?
50  */
51 #include <sys/buf.h>
52 
53 struct g_vfs_softc {
54 	struct mtx	 sc_mtx;
55 	struct bufobj	*sc_bo;
56 	struct g_event	*sc_event;
57 	int		 sc_active;
58 	bool		 sc_orphaned;
59 	int		 sc_enxio_active;
60 	int		 sc_enxio_reported;
61 };
62 
63 static struct buf_ops __g_vfs_bufops = {
64 	.bop_name =	"GEOM_VFS",
65 	.bop_write =	bufwrite,
66 	.bop_strategy =	g_vfs_strategy,
67 	.bop_sync =	bufsync,
68 	.bop_bdflush =	bufbdflush
69 };
70 
71 struct buf_ops *g_vfs_bufops = &__g_vfs_bufops;
72 
73 static g_orphan_t g_vfs_orphan;
74 
75 static struct g_class g_vfs_class = {
76 	.name =		"VFS",
77 	.version =	G_VERSION,
78 	.orphan =	g_vfs_orphan,
79 };
80 
81 DECLARE_GEOM_CLASS(g_vfs_class, g_vfs);
82 
83 static void
84 g_vfs_destroy(void *arg, int flags __unused)
85 {
86 	struct g_consumer *cp;
87 
88 	g_topology_assert();
89 	cp = arg;
90 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
91 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
92 	g_detach(cp);
93 	if (cp->geom->softc == NULL)
94 		g_wither_geom(cp->geom, ENXIO);
95 }
96 
97 static void
98 g_vfs_done(struct bio *bip)
99 {
100 	struct g_consumer *cp;
101 	struct g_event *event;
102 	struct g_vfs_softc *sc;
103 	struct buf *bp;
104 	int destroy;
105 	struct mount *mp;
106 	struct vnode *vp;
107 	struct cdev *cdevp;
108 
109 	/*
110 	 * Collect statistics on synchronous and asynchronous read
111 	 * and write counts for disks that have associated filesystems.
112 	 */
113 	bp = bip->bio_caller2;
114 	vp = bp->b_vp;
115 	if (vp != NULL) {
116 		/*
117 		 * If not a disk vnode, use its associated mount point
118 		 * otherwise use the mountpoint associated with the disk.
119 		 */
120 		VI_LOCK(vp);
121 		if (vp->v_type != VCHR ||
122 		    (cdevp = vp->v_rdev) == NULL ||
123 		    cdevp->si_devsw == NULL ||
124 		    (cdevp->si_devsw->d_flags & D_DISK) == 0)
125 			mp = vp->v_mount;
126 		else
127 			mp = cdevp->si_mountpt;
128 		if (mp != NULL) {
129 			if (bp->b_iocmd == BIO_READ) {
130 				if (LK_HOLDER(bp->b_lock.lk_lock) == LK_KERNPROC)
131 					mp->mnt_stat.f_asyncreads++;
132 				else
133 					mp->mnt_stat.f_syncreads++;
134 			} else if (bp->b_iocmd == BIO_WRITE) {
135 				if (LK_HOLDER(bp->b_lock.lk_lock) == LK_KERNPROC)
136 					mp->mnt_stat.f_asyncwrites++;
137 				else
138 					mp->mnt_stat.f_syncwrites++;
139 			}
140 		}
141 		VI_UNLOCK(vp);
142 	}
143 
144 	cp = bip->bio_from;
145 	sc = cp->geom->softc;
146 	if (bip->bio_error != 0 && bip->bio_error != EOPNOTSUPP) {
147 		if ((bp->b_xflags & BX_CVTENXIO) != 0) {
148 			if (atomic_cmpset_int(&sc->sc_enxio_active, 0, 1))
149 				printf("g_vfs_done(): %s converting all errors to ENXIO\n",
150 				    bip->bio_to->name);
151 		}
152 		if (sc->sc_enxio_active)
153 			bip->bio_error = ENXIO;
154 		if (bip->bio_error != ENXIO ||
155 		    atomic_cmpset_int(&sc->sc_enxio_reported, 0, 1)) {
156 			g_print_bio("g_vfs_done():", bip, "error = %d%s",
157 			    bip->bio_error,
158 			    bip->bio_error != ENXIO ? "" :
159 			    " supressing further ENXIO");
160 		}
161 	}
162 	bp->b_error = bip->bio_error;
163 	bp->b_ioflags = bip->bio_flags;
164 	if (bip->bio_error)
165 		bp->b_ioflags |= BIO_ERROR;
166 	bp->b_resid = bp->b_bcount - bip->bio_completed;
167 	g_destroy_bio(bip);
168 
169 	mtx_lock(&sc->sc_mtx);
170 	destroy = ((--sc->sc_active) == 0 && sc->sc_orphaned);
171 	if (destroy) {
172 		event = sc->sc_event;
173 		sc->sc_event = NULL;
174 	} else
175 		event = NULL;
176 	mtx_unlock(&sc->sc_mtx);
177 	if (destroy)
178 		g_post_event_ep(g_vfs_destroy, cp, event, NULL);
179 
180 	bufdone(bp);
181 }
182 
183 void
184 g_vfs_strategy(struct bufobj *bo, struct buf *bp)
185 {
186 	struct g_vfs_softc *sc;
187 	struct g_consumer *cp;
188 	struct bio *bip;
189 
190 	cp = bo->bo_private;
191 	sc = cp->geom->softc;
192 
193 	/*
194 	 * If the provider has orphaned us, just return ENXIO.
195 	 */
196 	mtx_lock(&sc->sc_mtx);
197 	if (sc->sc_orphaned || sc->sc_enxio_active) {
198 		mtx_unlock(&sc->sc_mtx);
199 		bp->b_error = ENXIO;
200 		bp->b_ioflags |= BIO_ERROR;
201 		bufdone(bp);
202 		return;
203 	}
204 	sc->sc_active++;
205 	mtx_unlock(&sc->sc_mtx);
206 
207 	bip = g_alloc_bio();
208 	bip->bio_cmd = bp->b_iocmd;
209 	bip->bio_offset = bp->b_iooffset;
210 	bip->bio_length = bp->b_bcount;
211 	bdata2bio(bp, bip);
212 	if ((bp->b_flags & B_BARRIER) != 0) {
213 		bip->bio_flags |= BIO_ORDERED;
214 		bp->b_flags &= ~B_BARRIER;
215 	}
216 	if (bp->b_iocmd == BIO_SPEEDUP)
217 		bip->bio_flags |= bp->b_ioflags;
218 	bip->bio_done = g_vfs_done;
219 	bip->bio_caller2 = bp;
220 #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING)
221 	buf_track(bp, __func__);
222 	bip->bio_track_bp = bp;
223 #endif
224 	g_io_request(bip, cp);
225 }
226 
227 static void
228 g_vfs_orphan(struct g_consumer *cp)
229 {
230 	struct g_geom *gp;
231 	struct g_event *event;
232 	struct g_vfs_softc *sc;
233 	int destroy;
234 
235 	g_topology_assert();
236 
237 	gp = cp->geom;
238 	g_trace(G_T_TOPOLOGY, "g_vfs_orphan(%p(%s))", cp, gp->name);
239 	sc = gp->softc;
240 	if (sc == NULL)
241 		return;
242 	event = g_alloc_event(M_WAITOK);
243 	mtx_lock(&sc->sc_mtx);
244 	KASSERT(sc->sc_event == NULL, ("g_vfs %p already has an event", sc));
245 	sc->sc_orphaned = true;
246 	destroy = (sc->sc_active == 0);
247 	if (!destroy) {
248 		sc->sc_event = event;
249 		event = NULL;
250 	}
251 	mtx_unlock(&sc->sc_mtx);
252 	if (destroy) {
253 		g_free(event);
254 		g_vfs_destroy(cp, 0);
255 	}
256 
257 	/*
258 	 * Do not destroy the geom.  Filesystem will do that during unmount.
259 	 */
260 }
261 
262 int
263 g_vfs_open(struct vnode *vp, struct g_consumer **cpp, const char *fsname, int wr)
264 {
265 	struct g_geom *gp;
266 	struct g_provider *pp;
267 	struct g_consumer *cp;
268 	struct g_vfs_softc *sc;
269 	struct bufobj *bo;
270 	int error;
271 
272 	g_topology_assert();
273 
274 	*cpp = NULL;
275 	bo = &vp->v_bufobj;
276 	if (bo->bo_private != vp)
277 		return (EBUSY);
278 
279 	pp = g_dev_getprovider(vp->v_rdev);
280 	if (pp == NULL)
281 		return (ENOENT);
282 	gp = g_new_geomf(&g_vfs_class, "%s.%s", fsname, pp->name);
283 	sc = g_malloc(sizeof(*sc), M_WAITOK | M_ZERO);
284 	mtx_init(&sc->sc_mtx, "g_vfs", NULL, MTX_DEF);
285 	sc->sc_bo = bo;
286 	gp->softc = sc;
287 	cp = g_new_consumer(gp);
288 	error = g_attach(cp, pp);
289 	if (error) {
290 		g_wither_geom(gp, ENXIO);
291 		return (error);
292 	}
293 	error = g_access(cp, 1, wr, wr);
294 	if (error) {
295 		g_wither_geom(gp, ENXIO);
296 		return (error);
297 	}
298 	vnode_create_vobject(vp, pp->mediasize, curthread);
299 	*cpp = cp;
300 	cp->private = vp;
301 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
302 	bo->bo_ops = g_vfs_bufops;
303 	bo->bo_private = cp;
304 	bo->bo_bsize = pp->sectorsize;
305 
306 	return (error);
307 }
308 
309 void
310 g_vfs_close(struct g_consumer *cp)
311 {
312 	struct g_geom *gp;
313 	struct g_vfs_softc *sc;
314 
315 	g_topology_assert();
316 
317 	gp = cp->geom;
318 	sc = gp->softc;
319 	bufobj_invalbuf(sc->sc_bo, V_SAVE, 0, 0);
320 	sc->sc_bo->bo_private = cp->private;
321 	gp->softc = NULL;
322 	mtx_destroy(&sc->sc_mtx);
323 	if (!sc->sc_orphaned || cp->provider == NULL)
324 		g_wither_geom_close(gp, ENXIO);
325 	KASSERT(sc->sc_event == NULL, ("g_vfs %p event is non-NULL", sc));
326 	g_free(sc);
327 }
328