1/*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 2002 Poul-Henning Kamp
5 * Copyright (c) 2002 Networks Associates Technology, Inc.
6 * All rights reserved.
7 *
8 * This software was developed for the FreeBSD Project by Poul-Henning Kamp
9 * and NAI Labs, the Security Research Division of Network Associates, Inc.
10 * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
11 * DARPA CHATS research program.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 *    notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 *    notice, this list of conditions and the following disclaimer in the
20 *    documentation and/or other materials provided with the distribution.
21 * 3. The names of the authors may not be used to endorse or promote
22 *    products derived from this software without specific prior written
23 *    permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 */
37
38#include <sys/cdefs.h>
39__FBSDID("$FreeBSD$");
40
41#include <sys/param.h>
42#include <sys/systm.h>
43#include <sys/kernel.h>
44#include <sys/malloc.h>
45#include <sys/bio.h>
46#include <sys/sysctl.h>
47#include <sys/proc.h>
48#include <sys/lock.h>
49#include <sys/mutex.h>
50#include <sys/errno.h>
51#include <sys/sbuf.h>
52#include <geom/geom.h>
53#include <geom/geom_slice.h>
54#include <machine/stdarg.h>
55
56static g_access_t g_slice_access;
57static g_start_t g_slice_start;
58
59static struct g_slicer *
60g_slice_alloc(unsigned nslice, unsigned scsize)
61{
62	struct g_slicer *gsp;
63
64	gsp = g_malloc(sizeof *gsp, M_WAITOK | M_ZERO);
65	if (scsize > 0)
66		gsp->softc = g_malloc(scsize, M_WAITOK | M_ZERO);
67	else
68		gsp->softc = NULL;
69	gsp->slices = g_malloc(nslice * sizeof(struct g_slice),
70	    M_WAITOK | M_ZERO);
71	gsp->nslice = nslice;
72	return (gsp);
73}
74
75static void
76g_slice_free(struct g_geom *gp)
77{
78	struct g_slicer *gsp;
79
80	gsp = gp->softc;
81	gp->softc = NULL;
82
83	/*
84	 * We can get multiple spoiled events before wither-washer
85	 * detaches our consumer, so this can get called multiple
86	 * times.
87	 */
88	if (gsp == NULL)
89		return;
90	g_free(gsp->slices);
91	if (gsp->hotspot != NULL)
92		g_free(gsp->hotspot);
93	if (gsp->softc != NULL)
94		g_free(gsp->softc);
95	g_free(gsp);
96}
97
98static int
99g_slice_access(struct g_provider *pp, int dr, int dw, int de)
100{
101	int error;
102	u_int u;
103	struct g_geom *gp;
104	struct g_consumer *cp;
105	struct g_provider *pp2;
106	struct g_slicer *gsp;
107	struct g_slice *gsl, *gsl2;
108
109	gp = pp->geom;
110	cp = LIST_FIRST(&gp->consumer);
111	KASSERT (cp != NULL, ("g_slice_access but no consumer"));
112	gsp = gp->softc;
113	if (dr > 0 || dw > 0 || de > 0) {
114		gsl = &gsp->slices[pp->index];
115		for (u = 0; u < gsp->nslice; u++) {
116			gsl2 = &gsp->slices[u];
117			if (gsl2->length == 0)
118				continue;
119			if (u == pp->index)
120				continue;
121			if (gsl->offset + gsl->length <= gsl2->offset)
122				continue;
123			if (gsl2->offset + gsl2->length <= gsl->offset)
124				continue;
125			/* overlap */
126			pp2 = gsl2->provider;
127			if ((pp->acw + dw) > 0 && pp2->ace > 0)
128				return (EPERM);
129			if ((pp->ace + de) > 0 && pp2->acw > 0)
130				return (EPERM);
131		}
132	}
133	/* On first open, grab an extra "exclusive" bit */
134	if (cp->acr == 0 && cp->acw == 0 && cp->ace == 0)
135		de++;
136	/* ... and let go of it on last close */
137	if ((cp->acr + dr) == 0 && (cp->acw + dw) == 0 && (cp->ace + de) == 1)
138		de--;
139	error = g_access(cp, dr, dw, de);
140
141	/*
142	 * Free the softc if all providers have been closed and this geom
143	 * is being removed.
144	 */
145	if (error == 0 && (gp->flags & G_GEOM_WITHER) != 0 &&
146	    (cp->acr + cp->acw + cp->ace) == 0)
147		g_slice_free(gp);
148
149	return (error);
150}
151
152/*
153 * XXX: It should be possible to specify here if we should finish all of the
154 * XXX: bio, or only the non-hot bits.  This would get messy if there were
155 * XXX: two hot spots in the same bio, so for now we simply finish off the
156 * XXX: entire bio.  Modifying hot data on the way to disk is frowned on
157 * XXX: so making that considerably harder is not a bad idea anyway.
158 */
159void
160g_slice_finish_hot(struct bio *bp)
161{
162	struct bio *bp2;
163	struct g_geom *gp;
164	struct g_consumer *cp;
165	struct g_slicer *gsp;
166	struct g_slice *gsl;
167	int idx;
168
169	KASSERT(bp->bio_to != NULL,
170	    ("NULL bio_to in g_slice_finish_hot(%p)", bp));
171	KASSERT(bp->bio_from != NULL,
172	    ("NULL bio_from in g_slice_finish_hot(%p)", bp));
173	gp = bp->bio_to->geom;
174	gsp = gp->softc;
175	cp = LIST_FIRST(&gp->consumer);
176	KASSERT(cp != NULL, ("NULL consumer in g_slice_finish_hot(%p)", bp));
177	idx = bp->bio_to->index;
178	gsl = &gsp->slices[idx];
179
180	bp2 = g_clone_bio(bp);
181	if (bp2 == NULL) {
182		g_io_deliver(bp, ENOMEM);
183		return;
184	}
185	if (bp2->bio_offset + bp2->bio_length > gsl->length)
186		bp2->bio_length = gsl->length - bp2->bio_offset;
187	bp2->bio_done = g_std_done;
188	bp2->bio_offset += gsl->offset;
189	g_io_request(bp2, cp);
190	return;
191}
192
193static void
194g_slice_done(struct bio *bp)
195{
196
197	KASSERT(bp->bio_cmd == BIO_GETATTR &&
198	    strcmp(bp->bio_attribute, "GEOM::ident") == 0,
199	    ("bio_cmd=0x%x bio_attribute=%s", bp->bio_cmd, bp->bio_attribute));
200
201	if (bp->bio_error == 0 && bp->bio_data[0] != '\0') {
202		char idx[8];
203
204		/* Add index to the ident received. */
205		snprintf(idx, sizeof(idx), "s%d",
206		    bp->bio_parent->bio_to->index);
207		if (strlcat(bp->bio_data, idx, bp->bio_length) >=
208		    bp->bio_length) {
209			bp->bio_error = EFAULT;
210		}
211	}
212	g_std_done(bp);
213}
214
215static void
216g_slice_start(struct bio *bp)
217{
218	struct bio *bp2;
219	struct g_provider *pp;
220	struct g_geom *gp;
221	struct g_consumer *cp;
222	struct g_slicer *gsp;
223	struct g_slice *gsl;
224	struct g_slice_hot *ghp;
225	int idx, error;
226	u_int m_index;
227	off_t t;
228
229	pp = bp->bio_to;
230	gp = pp->geom;
231	gsp = gp->softc;
232	cp = LIST_FIRST(&gp->consumer);
233	idx = pp->index;
234	gsl = &gsp->slices[idx];
235	switch(bp->bio_cmd) {
236	case BIO_READ:
237	case BIO_WRITE:
238	case BIO_DELETE:
239		if (bp->bio_offset > gsl->length) {
240			g_io_deliver(bp, EINVAL); /* XXX: EWHAT ? */
241			return;
242		}
243		/*
244		 * Check if we collide with any hot spaces, and call the
245		 * method once if so.
246		 */
247		t = bp->bio_offset + gsl->offset;
248		for (m_index = 0; m_index < gsp->nhotspot; m_index++) {
249			ghp = &gsp->hotspot[m_index];
250			if (t >= ghp->offset + ghp->length)
251				continue;
252			if (t + bp->bio_length <= ghp->offset)
253				continue;
254			switch(bp->bio_cmd) {
255			case BIO_READ:		idx = ghp->ract; break;
256			case BIO_WRITE:		idx = ghp->wact; break;
257			case BIO_DELETE:	idx = ghp->dact; break;
258			}
259			switch(idx) {
260			case G_SLICE_HOT_ALLOW:
261				/* Fall out and continue normal processing */
262				continue;
263			case G_SLICE_HOT_DENY:
264				g_io_deliver(bp, EROFS);
265				return;
266			case G_SLICE_HOT_START:
267				error = gsp->start(bp);
268				if (error && error != EJUSTRETURN)
269					g_io_deliver(bp, error);
270				return;
271			case G_SLICE_HOT_CALL:
272				error = g_post_event(gsp->hot, bp, M_NOWAIT,
273				    gp, NULL);
274				if (error)
275					g_io_deliver(bp, error);
276				return;
277			}
278			break;
279		}
280		bp2 = g_clone_bio(bp);
281		if (bp2 == NULL) {
282			g_io_deliver(bp, ENOMEM);
283			return;
284		}
285		if (bp2->bio_offset + bp2->bio_length > gsl->length)
286			bp2->bio_length = gsl->length - bp2->bio_offset;
287		bp2->bio_done = g_std_done;
288		bp2->bio_offset += gsl->offset;
289		g_io_request(bp2, cp);
290		return;
291	case BIO_GETATTR:
292		/* Give the real method a chance to override */
293		if (gsp->start != NULL && gsp->start(bp))
294			return;
295		if (!strcmp("GEOM::ident", bp->bio_attribute)) {
296			bp2 = g_clone_bio(bp);
297			if (bp2 == NULL) {
298				g_io_deliver(bp, ENOMEM);
299				return;
300			}
301			bp2->bio_done = g_slice_done;
302			g_io_request(bp2, cp);
303			return;
304		}
305		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
306			struct g_kerneldump *gkd;
307
308			gkd = (struct g_kerneldump *)bp->bio_data;
309			gkd->offset += gsp->slices[idx].offset;
310			if (gkd->length > gsp->slices[idx].length)
311				gkd->length = gsp->slices[idx].length;
312			/* now, pass it on downwards... */
313		}
314		/* FALLTHROUGH */
315	case BIO_SPEEDUP:
316	case BIO_FLUSH:
317		bp2 = g_clone_bio(bp);
318		if (bp2 == NULL) {
319			g_io_deliver(bp, ENOMEM);
320			return;
321		}
322		bp2->bio_done = g_std_done;
323		g_io_request(bp2, cp);
324		break;
325	default:
326		g_io_deliver(bp, EOPNOTSUPP);
327		return;
328	}
329}
330
331void
332g_slice_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp)
333{
334	struct g_slicer *gsp;
335
336	gsp = gp->softc;
337	if (indent == NULL) {
338		sbuf_printf(sb, " i %u", pp->index);
339		sbuf_printf(sb, " o %ju",
340		    (uintmax_t)gsp->slices[pp->index].offset);
341		return;
342	}
343	if (pp != NULL) {
344		sbuf_printf(sb, "%s<index>%u</index>\n", indent, pp->index);
345		sbuf_printf(sb, "%s<length>%ju</length>\n",
346		    indent, (uintmax_t)gsp->slices[pp->index].length);
347		sbuf_printf(sb, "%s<seclength>%ju</seclength>\n", indent,
348		    (uintmax_t)gsp->slices[pp->index].length / 512);
349		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
350		    (uintmax_t)gsp->slices[pp->index].offset);
351		sbuf_printf(sb, "%s<secoffset>%ju</secoffset>\n", indent,
352		    (uintmax_t)gsp->slices[pp->index].offset / 512);
353	}
354}
355
356int
357g_slice_config(struct g_geom *gp, u_int idx, int how, off_t offset, off_t length, u_int sectorsize, const char *fmt, ...)
358{
359	struct g_provider *pp, *pp2;
360	struct g_slicer *gsp;
361	struct g_slice *gsl;
362	va_list ap;
363	struct sbuf *sb;
364	int acc;
365
366	g_trace(G_T_TOPOLOGY, "g_slice_config(%s, %d, %d)",
367	     gp->name, idx, how);
368	g_topology_assert();
369	gsp = gp->softc;
370	if (idx >= gsp->nslice)
371		return(EINVAL);
372	gsl = &gsp->slices[idx];
373	pp = gsl->provider;
374	if (pp != NULL)
375		acc = pp->acr + pp->acw + pp->ace;
376	else
377		acc = 0;
378	if (acc != 0 && how != G_SLICE_CONFIG_FORCE) {
379		if (length < gsl->length)
380			return(EBUSY);
381		if (offset != gsl->offset)
382			return(EBUSY);
383	}
384	/* XXX: check offset + length <= MEDIASIZE */
385	if (how == G_SLICE_CONFIG_CHECK)
386		return (0);
387	gsl->length = length;
388	gsl->offset = offset;
389	gsl->sectorsize = sectorsize;
390	if (length == 0) {
391		if (pp == NULL)
392			return (0);
393		if (bootverbose)
394			printf("GEOM: Deconfigure %s\n", pp->name);
395		g_wither_provider(pp, ENXIO);
396		gsl->provider = NULL;
397		gsp->nprovider--;
398		return (0);
399	}
400	if (pp != NULL) {
401		if (bootverbose)
402			printf("GEOM: Reconfigure %s, start %jd length %jd end %jd\n",
403			    pp->name, (intmax_t)offset, (intmax_t)length,
404			    (intmax_t)(offset + length - 1));
405		g_resize_provider(pp, gsl->length);
406		return (0);
407	}
408	sb = sbuf_new_auto();
409	va_start(ap, fmt);
410	sbuf_vprintf(sb, fmt, ap);
411	va_end(ap);
412	sbuf_finish(sb);
413	pp = g_new_providerf(gp, "%s", sbuf_data(sb));
414	pp2 = LIST_FIRST(&gp->consumer)->provider;
415	pp->stripesize = pp2->stripesize;
416	pp->stripeoffset = pp2->stripeoffset + offset;
417	if (pp->stripesize > 0)
418		pp->stripeoffset %= pp->stripesize;
419	if (gsp->nhotspot == 0) {
420		pp->flags |= pp2->flags & G_PF_ACCEPT_UNMAPPED;
421		pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
422	}
423	if (0 && bootverbose)
424		printf("GEOM: Configure %s, start %jd length %jd end %jd\n",
425		    pp->name, (intmax_t)offset, (intmax_t)length,
426		    (intmax_t)(offset + length - 1));
427	pp->index = idx;
428	pp->mediasize = gsl->length;
429	pp->sectorsize = gsl->sectorsize;
430	gsl->provider = pp;
431	gsp->nprovider++;
432	g_error_provider(pp, 0);
433	sbuf_delete(sb);
434	return(0);
435}
436
437/*
438 * Configure "hotspots".  A hotspot is a piece of the parent device which
439 * this particular slicer cares about for some reason.  Typically because
440 * it contains meta-data used to configure the slicer.
441 * A hotspot is identified by its index number. The offset and length are
442 * relative to the parent device, and the three "?act" fields specify
443 * what action to take on BIO_READ, BIO_DELETE and BIO_WRITE.
444 *
445 * XXX: There may be a race relative to g_slice_start() here, if an existing
446 * XXX: hotspot is changed wile I/O is happening.  Should this become a problem
447 * XXX: we can protect the hotspot stuff with a mutex.
448 */
449
450int
451g_slice_conf_hot(struct g_geom *gp, u_int idx, off_t offset, off_t length, int ract, int dact, int wact)
452{
453	struct g_slicer *gsp;
454	struct g_slice_hot *gsl, *gsl2;
455	struct g_consumer *cp;
456	struct g_provider *pp;
457
458	g_trace(G_T_TOPOLOGY, "g_slice_conf_hot(%s, idx: %d, off: %jd, len: %jd)",
459	    gp->name, idx, (intmax_t)offset, (intmax_t)length);
460	g_topology_assert();
461	gsp = gp->softc;
462	/* Deny unmapped I/O and direct dispatch if hotspots are used. */
463	if (gsp->nhotspot == 0) {
464		LIST_FOREACH(pp, &gp->provider, provider)
465			pp->flags &= ~(G_PF_ACCEPT_UNMAPPED |
466			    G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE);
467		LIST_FOREACH(cp, &gp->consumer, consumer)
468			cp->flags &= ~(G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE);
469	}
470	gsl = gsp->hotspot;
471	if(idx >= gsp->nhotspot) {
472		gsl2 = g_malloc((idx + 1) * sizeof *gsl2, M_WAITOK | M_ZERO);
473		if (gsp->hotspot != NULL)
474			bcopy(gsp->hotspot, gsl2, gsp->nhotspot * sizeof *gsl2);
475		gsp->hotspot = gsl2;
476		if (gsp->hotspot != NULL)
477			g_free(gsl);
478		gsl = gsl2;
479		gsp->nhotspot = idx + 1;
480	}
481	gsl[idx].offset = offset;
482	gsl[idx].length = length;
483	KASSERT(!((ract | dact | wact) & G_SLICE_HOT_START)
484	    || gsp->start != NULL, ("G_SLICE_HOT_START but no slice->start"));
485	/* XXX: check that we _have_ a start function if HOT_START specified */
486	gsl[idx].ract = ract;
487	gsl[idx].dact = dact;
488	gsl[idx].wact = wact;
489	return (0);
490}
491
492void
493g_slice_orphan(struct g_consumer *cp)
494{
495	struct g_geom *gp;
496
497	g_topology_assert();
498	gp = cp->geom;
499	g_trace(G_T_TOPOLOGY, "%s(%p/%s)", __func__, cp, gp->name);
500	g_wither_geom(gp, ENXIO);
501
502	/*
503	 * We can safely free the softc now if there are no accesses,
504	 * otherwise g_slice_access() will do that after the last close.
505	 */
506	if ((cp->acr + cp->acw + cp->ace) == 0)
507		g_slice_free(gp);
508}
509
510void
511g_slice_spoiled(struct g_consumer *cp)
512{
513
514	g_trace(G_T_TOPOLOGY, "%s(%p/%s)", __func__, cp, cp->geom->name);
515	cp->flags |= G_CF_ORPHAN;
516	g_slice_orphan(cp);
517}
518
519int
520g_slice_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
521{
522
523	g_slice_spoiled(LIST_FIRST(&gp->consumer));
524	return (0);
525}
526
527struct g_geom *
528g_slice_new(struct g_class *mp, u_int slices, struct g_provider *pp, struct g_consumer **cpp, void *extrap, int extra, g_slice_start_t *start)
529{
530	struct g_geom *gp;
531	struct g_slicer *gsp;
532	struct g_consumer *cp;
533	void **vp;
534	int error;
535
536	g_topology_assert();
537	vp = (void **)extrap;
538	gp = g_new_geomf(mp, "%s", pp->name);
539	gsp = g_slice_alloc(slices, extra);
540	gsp->start = start;
541	gp->softc = gsp;
542	gp->start = g_slice_start;
543	gp->access = g_slice_access;
544	gp->orphan = g_slice_orphan;
545	gp->spoiled = g_slice_spoiled;
546	if (gp->dumpconf == NULL)
547		gp->dumpconf = g_slice_dumpconf;
548	if (gp->class->destroy_geom == NULL)
549		gp->class->destroy_geom = g_slice_destroy_geom;
550	cp = g_new_consumer(gp);
551	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
552	error = g_attach(cp, pp);
553	if (error == 0)
554		error = g_access(cp, 1, 0, 0);
555	if (error) {
556		g_wither_geom(gp, ENXIO);
557		return (NULL);
558	}
559	if (extrap != NULL)
560		*vp = gsp->softc;
561	*cpp = cp;
562	return (gp);
563}
564