1/*-
2 * Copyright (c) 2004 Max Khon
3 * Copyright (c) 2014 Juniper Networks, Inc.
4 * Copyright (c) 2006-2016 Maxim Sobolev <sobomax@FreeBSD.org>
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: stable/11/sys/geom/uzip/g_uzip.c 356579 2020-01-10 00:43:08Z mav $");
31
32#include <sys/param.h>
33#include <sys/bio.h>
34#include <sys/endian.h>
35#include <sys/errno.h>
36#include <sys/kernel.h>
37#include <sys/lock.h>
38#include <sys/mutex.h>
39#include <sys/malloc.h>
40#include <sys/sysctl.h>
41#include <sys/systm.h>
42#include <sys/kthread.h>
43
44#include <geom/geom.h>
45
46#include <geom/uzip/g_uzip.h>
47#include <geom/uzip/g_uzip_cloop.h>
48#include <geom/uzip/g_uzip_softc.h>
49#include <geom/uzip/g_uzip_dapi.h>
50#include <geom/uzip/g_uzip_zlib.h>
51#include <geom/uzip/g_uzip_lzma.h>
52#include <geom/uzip/g_uzip_wrkthr.h>
53
54#include "opt_geom.h"
55
56MALLOC_DEFINE(M_GEOM_UZIP, "geom_uzip", "GEOM UZIP data structures");
57
58FEATURE(geom_uzip, "GEOM read-only compressed disks support");
59
60struct g_uzip_blk {
61        uint64_t offset;
62        uint32_t blen;
63        unsigned char last:1;
64        unsigned char padded:1;
65#define BLEN_UNDEF      UINT32_MAX
66};
67
68#ifndef ABS
69#define	ABS(a)			((a) < 0 ? -(a) : (a))
70#endif
71
72#define BLK_IN_RANGE(mcn, bcn, ilen)	\
73    (((bcn) != BLEN_UNDEF) && ( \
74	((ilen) >= 0 && (mcn >= bcn) && (mcn <= ((intmax_t)(bcn) + (ilen)))) || \
75	((ilen) < 0 && (mcn <= bcn) && (mcn >= ((intmax_t)(bcn) + (ilen)))) \
76    ))
77
78#ifdef GEOM_UZIP_DEBUG
79# define GEOM_UZIP_DBG_DEFAULT	3
80#else
81# define GEOM_UZIP_DBG_DEFAULT	0
82#endif
83
84#define	GUZ_DBG_ERR	1
85#define	GUZ_DBG_INFO	2
86#define	GUZ_DBG_IO	3
87#define	GUZ_DBG_TOC	4
88
89#define	GUZ_DEV_SUFX	".uzip"
90#define	GUZ_DEV_NAME(p)	(p GUZ_DEV_SUFX)
91
92static char g_uzip_attach_to[MAXPATHLEN] = {"*"};
93static char g_uzip_noattach_to[MAXPATHLEN] = {GUZ_DEV_NAME("*")};
94TUNABLE_STR("kern.geom.uzip.attach_to", g_uzip_attach_to,
95    sizeof(g_uzip_attach_to));
96TUNABLE_STR("kern.geom.uzip.noattach_to", g_uzip_noattach_to,
97    sizeof(g_uzip_noattach_to));
98
99SYSCTL_DECL(_kern_geom);
100SYSCTL_NODE(_kern_geom, OID_AUTO, uzip, CTLFLAG_RW, 0, "GEOM_UZIP stuff");
101static u_int g_uzip_debug = GEOM_UZIP_DBG_DEFAULT;
102SYSCTL_UINT(_kern_geom_uzip, OID_AUTO, debug, CTLFLAG_RWTUN, &g_uzip_debug, 0,
103    "Debug level (0-4)");
104static u_int g_uzip_debug_block = BLEN_UNDEF;
105SYSCTL_UINT(_kern_geom_uzip, OID_AUTO, debug_block, CTLFLAG_RWTUN,
106    &g_uzip_debug_block, 0, "Debug operations around specific cluster#");
107
108#define	DPRINTF(lvl, a)		\
109	if ((lvl) <= g_uzip_debug) { \
110		printf a; \
111	}
112#define	DPRINTF_BLK(lvl, cn, a)	\
113	if ((lvl) <= g_uzip_debug || \
114	    BLK_IN_RANGE(cn, g_uzip_debug_block, 8) || \
115	    BLK_IN_RANGE(cn, g_uzip_debug_block, -8)) { \
116		printf a; \
117	}
118#define	DPRINTF_BRNG(lvl, bcn, ecn, a) \
119	KASSERT(bcn < ecn, ("DPRINTF_BRNG: invalid range (%ju, %ju)", \
120	    (uintmax_t)bcn, (uintmax_t)ecn)); \
121	if (((lvl) <= g_uzip_debug) || \
122	    BLK_IN_RANGE(g_uzip_debug_block, bcn, \
123	     (intmax_t)ecn - (intmax_t)bcn)) { \
124		printf a; \
125	}
126
127#define	UZIP_CLASS_NAME	"UZIP"
128
129/*
130 * Maximum allowed valid block size (to prevent foot-shooting)
131 */
132#define	MAX_BLKSZ	(MAXPHYS)
133
134static char CLOOP_MAGIC_START[] = "#!/bin/sh\n";
135
136static void g_uzip_read_done(struct bio *bp);
137static void g_uzip_do(struct g_uzip_softc *, struct bio *bp);
138
139static void
140g_uzip_softc_free(struct g_geom *gp)
141{
142	struct g_uzip_softc *sc = gp->softc;
143
144	DPRINTF(GUZ_DBG_INFO, ("%s: %d requests, %d cached\n",
145	    gp->name, sc->req_total, sc->req_cached));
146
147	mtx_lock(&sc->queue_mtx);
148	sc->wrkthr_flags |= GUZ_SHUTDOWN;
149	wakeup(sc);
150	while (!(sc->wrkthr_flags & GUZ_EXITING)) {
151		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "guzfree",
152		    hz / 10);
153	}
154	mtx_unlock(&sc->queue_mtx);
155
156	sc->dcp->free(sc->dcp);
157	free(sc->toc, M_GEOM_UZIP);
158	mtx_destroy(&sc->queue_mtx);
159	mtx_destroy(&sc->last_mtx);
160	free(sc->last_buf, M_GEOM_UZIP);
161	free(sc, M_GEOM_UZIP);
162	gp->softc = NULL;
163}
164
165static int
166g_uzip_cached(struct g_geom *gp, struct bio *bp)
167{
168	struct g_uzip_softc *sc;
169	off_t ofs;
170	size_t blk, blkofs, usz;
171
172	sc = gp->softc;
173	ofs = bp->bio_offset + bp->bio_completed;
174	blk = ofs / sc->blksz;
175	mtx_lock(&sc->last_mtx);
176	if (blk == sc->last_blk) {
177		blkofs = ofs % sc->blksz;
178		usz = sc->blksz - blkofs;
179		if (bp->bio_resid < usz)
180			usz = bp->bio_resid;
181		memcpy(bp->bio_data + bp->bio_completed, sc->last_buf + blkofs,
182		    usz);
183		sc->req_cached++;
184		mtx_unlock(&sc->last_mtx);
185
186		DPRINTF(GUZ_DBG_IO, ("%s/%s: %p: offset=%jd: got %jd bytes "
187		    "from cache\n", __func__, gp->name, bp, (intmax_t)ofs,
188		    (intmax_t)usz));
189
190		bp->bio_completed += usz;
191		bp->bio_resid -= usz;
192
193		if (bp->bio_resid == 0) {
194			g_io_deliver(bp, 0);
195			return (1);
196		}
197	} else
198		mtx_unlock(&sc->last_mtx);
199
200	return (0);
201}
202
203#define BLK_ENDS(sc, bi)	((sc)->toc[(bi)].offset + \
204    (sc)->toc[(bi)].blen)
205
206#define BLK_IS_CONT(sc, bi)	(BLK_ENDS((sc), (bi) - 1) == \
207    (sc)->toc[(bi)].offset)
208#define	BLK_IS_NIL(sc, bi)	((sc)->toc[(bi)].blen == 0)
209
210#define TOFF_2_BOFF(sc, pp, bi)	    ((sc)->toc[(bi)].offset - \
211    (sc)->toc[(bi)].offset % (pp)->sectorsize)
212#define	TLEN_2_BLEN(sc, pp, bp, ei) roundup(BLK_ENDS((sc), (ei)) - \
213    (bp)->bio_offset, (pp)->sectorsize)
214
215static int
216g_uzip_request(struct g_geom *gp, struct bio *bp)
217{
218	struct g_uzip_softc *sc;
219	struct bio *bp2;
220	struct g_consumer *cp;
221	struct g_provider *pp;
222	off_t ofs, start_blk_ofs;
223	size_t i, start_blk, end_blk, zsize;
224
225	if (g_uzip_cached(gp, bp) != 0)
226		return (1);
227
228	sc = gp->softc;
229
230	cp = LIST_FIRST(&gp->consumer);
231	pp = cp->provider;
232
233	ofs = bp->bio_offset + bp->bio_completed;
234	start_blk = ofs / sc->blksz;
235	KASSERT(start_blk < sc->nblocks, ("start_blk out of range"));
236	end_blk = howmany(ofs + bp->bio_resid, sc->blksz);
237	KASSERT(end_blk <= sc->nblocks, ("end_blk out of range"));
238
239	for (; BLK_IS_NIL(sc, start_blk) && start_blk < end_blk; start_blk++) {
240		/* Fill in any leading Nil blocks */
241		start_blk_ofs = ofs % sc->blksz;
242		zsize = MIN(sc->blksz - start_blk_ofs, bp->bio_resid);
243		DPRINTF_BLK(GUZ_DBG_IO, start_blk, ("%s/%s: %p/%ju: "
244		    "filling %ju zero bytes\n", __func__, gp->name, gp,
245		    (uintmax_t)bp->bio_completed, (uintmax_t)zsize));
246		bzero(bp->bio_data + bp->bio_completed, zsize);
247		bp->bio_completed += zsize;
248		bp->bio_resid -= zsize;
249		ofs += zsize;
250	}
251
252	if (start_blk == end_blk) {
253		KASSERT(bp->bio_resid == 0, ("bp->bio_resid is invalid"));
254		/*
255		 * No non-Nil data is left, complete request immediately.
256		 */
257		DPRINTF(GUZ_DBG_IO, ("%s/%s: %p: all done returning %ju "
258		    "bytes\n", __func__, gp->name, gp,
259		    (uintmax_t)bp->bio_completed));
260		g_io_deliver(bp, 0);
261		return (1);
262	}
263
264	for (i = start_blk + 1; i < end_blk; i++) {
265		/* Trim discontinuous areas if any */
266		if (!BLK_IS_CONT(sc, i)) {
267			end_blk = i;
268			break;
269		}
270	}
271
272	DPRINTF_BRNG(GUZ_DBG_IO, start_blk, end_blk, ("%s/%s: %p: "
273	    "start=%u (%ju[%jd]), end=%u (%ju)\n", __func__, gp->name, bp,
274	    (u_int)start_blk, (uintmax_t)sc->toc[start_blk].offset,
275	    (intmax_t)sc->toc[start_blk].blen,
276	    (u_int)end_blk, (uintmax_t)BLK_ENDS(sc, end_blk - 1)));
277
278	bp2 = g_clone_bio(bp);
279	if (bp2 == NULL) {
280		g_io_deliver(bp, ENOMEM);
281		return (1);
282	}
283	bp2->bio_done = g_uzip_read_done;
284
285	bp2->bio_offset = TOFF_2_BOFF(sc, pp, start_blk);
286	while (1) {
287		bp2->bio_length = TLEN_2_BLEN(sc, pp, bp2, end_blk - 1);
288		if (bp2->bio_length <= MAXPHYS) {
289			break;
290		}
291		if (end_blk == (start_blk + 1)) {
292			break;
293		}
294		end_blk--;
295	}
296
297	DPRINTF(GUZ_DBG_IO, ("%s/%s: bp2->bio_length = %jd, "
298	    "bp2->bio_offset = %jd\n", __func__, gp->name,
299	    (intmax_t)bp2->bio_length, (intmax_t)bp2->bio_offset));
300
301	bp2->bio_data = malloc(bp2->bio_length, M_GEOM_UZIP, M_NOWAIT);
302	if (bp2->bio_data == NULL) {
303		g_destroy_bio(bp2);
304		g_io_deliver(bp, ENOMEM);
305		return (1);
306	}
307
308	DPRINTF_BRNG(GUZ_DBG_IO, start_blk, end_blk, ("%s/%s: %p: "
309	    "reading %jd bytes from offset %jd\n", __func__, gp->name, bp,
310	    (intmax_t)bp2->bio_length, (intmax_t)bp2->bio_offset));
311
312	g_io_request(bp2, cp);
313	return (0);
314}
315
316static void
317g_uzip_read_done(struct bio *bp)
318{
319	struct bio *bp2;
320	struct g_geom *gp;
321	struct g_uzip_softc *sc;
322
323	bp2 = bp->bio_parent;
324	gp = bp2->bio_to->geom;
325	sc = gp->softc;
326
327	mtx_lock(&sc->queue_mtx);
328	bioq_disksort(&sc->bio_queue, bp);
329	mtx_unlock(&sc->queue_mtx);
330	wakeup(sc);
331}
332
333static int
334g_uzip_memvcmp(const void *memory, unsigned char val, size_t size)
335{
336	const u_char *mm;
337
338	mm = (const u_char *)memory;
339	return (*mm == val) && memcmp(mm, mm + 1, size - 1) == 0;
340}
341
342static void
343g_uzip_do(struct g_uzip_softc *sc, struct bio *bp)
344{
345	struct bio *bp2;
346	struct g_provider *pp;
347	struct g_consumer *cp;
348	struct g_geom *gp;
349	char *data, *data2;
350	off_t ofs;
351	size_t blk, blkofs, len, ulen, firstblk;
352	int err;
353
354	bp2 = bp->bio_parent;
355	gp = bp2->bio_to->geom;
356
357	cp = LIST_FIRST(&gp->consumer);
358	pp = cp->provider;
359
360	bp2->bio_error = bp->bio_error;
361	if (bp2->bio_error != 0)
362		goto done;
363
364	/* Make sure there's forward progress. */
365	if (bp->bio_completed == 0) {
366		bp2->bio_error = ECANCELED;
367		goto done;
368	}
369
370	ofs = bp2->bio_offset + bp2->bio_completed;
371	firstblk = blk = ofs / sc->blksz;
372	blkofs = ofs % sc->blksz;
373	data = bp->bio_data + sc->toc[blk].offset % pp->sectorsize;
374	data2 = bp2->bio_data + bp2->bio_completed;
375	while (bp->bio_completed && bp2->bio_resid) {
376		if (blk > firstblk && !BLK_IS_CONT(sc, blk)) {
377			DPRINTF_BLK(GUZ_DBG_IO, blk, ("%s/%s: %p: backref'ed "
378			    "cluster #%u requested, looping around\n",
379			    __func__, gp->name, bp2, (u_int)blk));
380			goto done;
381		}
382		ulen = MIN(sc->blksz - blkofs, bp2->bio_resid);
383		len = sc->toc[blk].blen;
384		DPRINTF(GUZ_DBG_IO, ("%s/%s: %p/%ju: data2=%p, ulen=%u, "
385		    "data=%p, len=%u\n", __func__, gp->name, gp,
386		    bp->bio_completed, data2, (u_int)ulen, data, (u_int)len));
387		if (len == 0) {
388			/* All zero block: no cache update */
389zero_block:
390			bzero(data2, ulen);
391		} else if (len <= bp->bio_completed) {
392			mtx_lock(&sc->last_mtx);
393			err = sc->dcp->decompress(sc->dcp, gp->name, data,
394			    len, sc->last_buf);
395			if (err != 0 && sc->toc[blk].last != 0) {
396				/*
397				 * Last block decompression has failed, check
398				 * if it's just zero padding.
399				 */
400				if (g_uzip_memvcmp(data, '\0', len) == 0) {
401					sc->toc[blk].blen = 0;
402					sc->last_blk = -1;
403					mtx_unlock(&sc->last_mtx);
404					len = 0;
405					goto zero_block;
406				}
407			}
408			if (err != 0) {
409				sc->last_blk = -1;
410				mtx_unlock(&sc->last_mtx);
411				bp2->bio_error = EILSEQ;
412				DPRINTF(GUZ_DBG_ERR, ("%s/%s: decompress"
413				    "(%p, %ju, %ju) failed\n", __func__,
414				    gp->name, sc->dcp, (uintmax_t)blk,
415				    (uintmax_t)len));
416				goto done;
417			}
418			sc->last_blk = blk;
419			memcpy(data2, sc->last_buf + blkofs, ulen);
420			mtx_unlock(&sc->last_mtx);
421			err = sc->dcp->rewind(sc->dcp, gp->name);
422			if (err != 0) {
423				bp2->bio_error = EILSEQ;
424				DPRINTF(GUZ_DBG_ERR, ("%s/%s: rewind(%p) "
425				    "failed\n", __func__, gp->name, sc->dcp));
426				goto done;
427			}
428			data += len;
429		} else
430			break;
431
432		data2 += ulen;
433		bp2->bio_completed += ulen;
434		bp2->bio_resid -= ulen;
435		bp->bio_completed -= len;
436		blkofs = 0;
437		blk++;
438	}
439
440done:
441	/* Finish processing the request. */
442	free(bp->bio_data, M_GEOM_UZIP);
443	g_destroy_bio(bp);
444	if (bp2->bio_error != 0 || bp2->bio_resid == 0)
445		g_io_deliver(bp2, bp2->bio_error);
446	else
447		g_uzip_request(gp, bp2);
448}
449
450static void
451g_uzip_start(struct bio *bp)
452{
453	struct g_provider *pp;
454	struct g_geom *gp;
455	struct g_uzip_softc *sc;
456
457	pp = bp->bio_to;
458	gp = pp->geom;
459
460	DPRINTF(GUZ_DBG_IO, ("%s/%s: %p: cmd=%d, offset=%jd, length=%jd, "
461	    "buffer=%p\n", __func__, gp->name, bp, bp->bio_cmd,
462	    (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length, bp->bio_data));
463
464	sc = gp->softc;
465	sc->req_total++;
466
467	if (bp->bio_cmd != BIO_READ) {
468		g_io_deliver(bp, EOPNOTSUPP);
469		return;
470	}
471
472	bp->bio_resid = bp->bio_length;
473	bp->bio_completed = 0;
474
475	g_uzip_request(gp, bp);
476}
477
478static void
479g_uzip_orphan(struct g_consumer *cp)
480{
481	struct g_geom *gp;
482
483	g_topology_assert();
484	G_VALID_CONSUMER(cp);
485	gp = cp->geom;
486	g_trace(G_T_TOPOLOGY, "%s(%p/%s)", __func__, cp, gp->name);
487	g_wither_geom(gp, ENXIO);
488
489	/*
490	 * We can safely free the softc now if there are no accesses,
491	 * otherwise g_uzip_access() will do that after the last close.
492	 */
493	if ((cp->acr + cp->acw + cp->ace) == 0)
494		g_uzip_softc_free(gp);
495}
496
497static void
498g_uzip_spoiled(struct g_consumer *cp)
499{
500
501	g_trace(G_T_TOPOLOGY, "%s(%p/%s)", __func__, cp, cp->geom->name);
502	cp->flags |= G_CF_ORPHAN;
503	g_uzip_orphan(cp);
504}
505
506static int
507g_uzip_access(struct g_provider *pp, int dr, int dw, int de)
508{
509	struct g_geom *gp;
510	struct g_consumer *cp;
511	int error;
512
513	gp = pp->geom;
514	cp = LIST_FIRST(&gp->consumer);
515	KASSERT (cp != NULL, ("g_uzip_access but no consumer"));
516
517	if (cp->acw + dw > 0)
518		return (EROFS);
519
520	error = g_access(cp, dr, dw, de);
521
522	/*
523	 * Free the softc if all providers have been closed and this geom
524	 * is being removed.
525	 */
526	if (error == 0 && (gp->flags & G_GEOM_WITHER) != 0 &&
527	    (cp->acr + cp->acw + cp->ace) == 0)
528		g_uzip_softc_free(gp);
529
530	return (error);
531}
532
533static int
534g_uzip_parse_toc(struct g_uzip_softc *sc, struct g_provider *pp,
535    struct g_geom *gp)
536{
537	uint32_t i, j, backref_to;
538	uint64_t max_offset, min_offset;
539	struct g_uzip_blk *last_blk;
540
541	min_offset = sizeof(struct cloop_header) +
542	    (sc->nblocks + 1) * sizeof(uint64_t);
543	max_offset = sc->toc[0].offset - 1;
544	last_blk = &sc->toc[0];
545	for (i = 0; i < sc->nblocks; i++) {
546		/* First do some bounds checking */
547		if ((sc->toc[i].offset < min_offset) ||
548		    (sc->toc[i].offset > pp->mediasize)) {
549			goto error_offset;
550		}
551		DPRINTF_BLK(GUZ_DBG_IO, i, ("%s: cluster #%u "
552		    "offset=%ju max_offset=%ju\n", gp->name,
553		    (u_int)i, (uintmax_t)sc->toc[i].offset,
554		    (uintmax_t)max_offset));
555		backref_to = BLEN_UNDEF;
556		if (sc->toc[i].offset < max_offset) {
557			/*
558			 * For the backref'ed blocks search already parsed
559			 * TOC entries for the matching offset and copy the
560			 * size from matched entry.
561			 */
562			for (j = 0; j <= i; j++) {
563                                if (sc->toc[j].offset == sc->toc[i].offset &&
564				    !BLK_IS_NIL(sc, j)) {
565                                        break;
566                                }
567                                if (j != i) {
568					continue;
569				}
570				DPRINTF(GUZ_DBG_ERR, ("%s: cannot match "
571				    "backref'ed offset at cluster #%u\n",
572				    gp->name, i));
573				return (-1);
574			}
575			sc->toc[i].blen = sc->toc[j].blen;
576			backref_to = j;
577		} else {
578			last_blk = &sc->toc[i];
579			/*
580			 * For the "normal blocks" seek forward until we hit
581			 * block whose offset is larger than ours and assume
582			 * it's going to be the next one.
583			 */
584			for (j = i + 1; j < sc->nblocks; j++) {
585				if (sc->toc[j].offset > max_offset) {
586					break;
587				}
588			}
589			sc->toc[i].blen = sc->toc[j].offset -
590			    sc->toc[i].offset;
591			if (BLK_ENDS(sc, i) > pp->mediasize) {
592				DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u "
593				    "extends past media boundary (%ju > %ju)\n",
594				    gp->name, (u_int)i,
595				    (uintmax_t)BLK_ENDS(sc, i),
596				    (intmax_t)pp->mediasize));
597				return (-1);
598			}
599			KASSERT(max_offset <= sc->toc[i].offset, (
600			    "%s: max_offset is incorrect: %ju",
601			    gp->name, (uintmax_t)max_offset));
602			max_offset = BLK_ENDS(sc, i) - 1;
603		}
604		DPRINTF_BLK(GUZ_DBG_TOC, i, ("%s: cluster #%u, original %u "
605		    "bytes, in %u bytes", gp->name, i, sc->blksz,
606		    sc->toc[i].blen));
607		if (backref_to != BLEN_UNDEF) {
608			DPRINTF_BLK(GUZ_DBG_TOC, i, (" (->#%u)",
609			    (u_int)backref_to));
610		}
611		DPRINTF_BLK(GUZ_DBG_TOC, i, ("\n"));
612	}
613	last_blk->last = 1;
614	/* Do a second pass to validate block lengths */
615	for (i = 0; i < sc->nblocks; i++) {
616		if (sc->toc[i].blen > sc->dcp->max_blen) {
617			if (sc->toc[i].last == 0) {
618				DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u "
619				    "length (%ju) exceeds "
620				    "max_blen (%ju)\n", gp->name, i,
621				    (uintmax_t)sc->toc[i].blen,
622				    (uintmax_t)sc->dcp->max_blen));
623				return (-1);
624			}
625			DPRINTF(GUZ_DBG_INFO, ("%s: cluster #%u extra "
626			    "padding is detected, trimmed to %ju\n",
627			    gp->name, i, (uintmax_t)sc->dcp->max_blen));
628			    sc->toc[i].blen = sc->dcp->max_blen;
629			sc->toc[i].padded = 1;
630		}
631	}
632	return (0);
633
634error_offset:
635	DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u: invalid offset %ju, "
636	    "min_offset=%ju mediasize=%jd\n", gp->name, (u_int)i,
637	    sc->toc[i].offset, min_offset, pp->mediasize));
638	return (-1);
639}
640
641static struct g_geom *
642g_uzip_taste(struct g_class *mp, struct g_provider *pp, int flags)
643{
644	int error;
645	uint32_t i, total_offsets, offsets_read, blk;
646	void *buf;
647	struct cloop_header *header;
648	struct g_consumer *cp;
649	struct g_geom *gp;
650	struct g_provider *pp2;
651	struct g_uzip_softc *sc;
652	enum {
653		G_UZIP = 1,
654		G_ULZMA
655	} type;
656
657	g_trace(G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name);
658	g_topology_assert();
659
660	/* Skip providers that are already open for writing. */
661	if (pp->acw > 0)
662		return (NULL);
663
664	if ((fnmatch(g_uzip_attach_to, pp->name, 0) != 0) ||
665	    (fnmatch(g_uzip_noattach_to, pp->name, 0) == 0)) {
666		DPRINTF(GUZ_DBG_INFO, ("%s(%s,%s), ignoring\n", __func__,
667		    mp->name, pp->name));
668		return (NULL);
669	}
670
671	buf = NULL;
672
673	/*
674	 * Create geom instance.
675	 */
676	gp = g_new_geomf(mp, GUZ_DEV_NAME("%s"), pp->name);
677	cp = g_new_consumer(gp);
678	error = g_attach(cp, pp);
679	if (error == 0)
680		error = g_access(cp, 1, 0, 0);
681	if (error) {
682		goto e1;
683	}
684	g_topology_unlock();
685
686	/*
687	 * Read cloop header, look for CLOOP magic, perform
688	 * other validity checks.
689	 */
690	DPRINTF(GUZ_DBG_INFO, ("%s: media sectorsize %u, mediasize %jd\n",
691	    gp->name, pp->sectorsize, (intmax_t)pp->mediasize));
692	buf = g_read_data(cp, 0, pp->sectorsize, NULL);
693	if (buf == NULL)
694		goto e2;
695	header = (struct cloop_header *) buf;
696	if (strncmp(header->magic, CLOOP_MAGIC_START,
697	    sizeof(CLOOP_MAGIC_START) - 1) != 0) {
698		DPRINTF(GUZ_DBG_ERR, ("%s: no CLOOP magic\n", gp->name));
699		goto e3;
700	}
701
702	switch (header->magic[CLOOP_OFS_COMPR]) {
703	case CLOOP_COMP_LZMA:
704	case CLOOP_COMP_LZMA_DDP:
705		type = G_ULZMA;
706		if (header->magic[CLOOP_OFS_VERSN] < CLOOP_MINVER_LZMA) {
707			DPRINTF(GUZ_DBG_ERR, ("%s: image version too old\n",
708			    gp->name));
709			goto e3;
710		}
711		DPRINTF(GUZ_DBG_INFO, ("%s: GEOM_UZIP_LZMA image found\n",
712		    gp->name));
713		break;
714	case CLOOP_COMP_LIBZ:
715	case CLOOP_COMP_LIBZ_DDP:
716		type = G_UZIP;
717		if (header->magic[CLOOP_OFS_VERSN] < CLOOP_MINVER_ZLIB) {
718			DPRINTF(GUZ_DBG_ERR, ("%s: image version too old\n",
719			    gp->name));
720			goto e3;
721		}
722		DPRINTF(GUZ_DBG_INFO, ("%s: GEOM_UZIP_ZLIB image found\n",
723		    gp->name));
724		break;
725	default:
726		DPRINTF(GUZ_DBG_ERR, ("%s: unsupported image type\n",
727		    gp->name));
728                goto e3;
729        }
730
731	/*
732	 * Initialize softc and read offsets.
733	 */
734	sc = malloc(sizeof(*sc), M_GEOM_UZIP, M_WAITOK | M_ZERO);
735	gp->softc = sc;
736	sc->blksz = ntohl(header->blksz);
737	sc->nblocks = ntohl(header->nblocks);
738	if (sc->blksz % 512 != 0) {
739		printf("%s: block size (%u) should be multiple of 512.\n",
740		    gp->name, sc->blksz);
741		goto e4;
742	}
743	if (sc->blksz > MAX_BLKSZ) {
744		printf("%s: block size (%u) should not be larger than %d.\n",
745		    gp->name, sc->blksz, MAX_BLKSZ);
746	}
747	total_offsets = sc->nblocks + 1;
748	if (sizeof(struct cloop_header) +
749	    total_offsets * sizeof(uint64_t) > pp->mediasize) {
750		printf("%s: media too small for %u blocks\n",
751		    gp->name, sc->nblocks);
752		goto e4;
753	}
754	sc->toc = malloc(total_offsets * sizeof(struct g_uzip_blk),
755	    M_GEOM_UZIP, M_WAITOK | M_ZERO);
756	offsets_read = MIN(total_offsets,
757	    (pp->sectorsize - sizeof(*header)) / sizeof(uint64_t));
758	for (i = 0; i < offsets_read; i++) {
759		sc->toc[i].offset = be64toh(((uint64_t *) (header + 1))[i]);
760		sc->toc[i].blen = BLEN_UNDEF;
761	}
762	DPRINTF(GUZ_DBG_INFO, ("%s: %u offsets in the first sector\n",
763	       gp->name, offsets_read));
764	for (blk = 1; offsets_read < total_offsets; blk++) {
765		uint32_t nread;
766
767		free(buf, M_GEOM);
768		buf = g_read_data(
769		    cp, blk * pp->sectorsize, pp->sectorsize, NULL);
770		if (buf == NULL)
771			goto e5;
772		nread = MIN(total_offsets - offsets_read,
773		     pp->sectorsize / sizeof(uint64_t));
774		DPRINTF(GUZ_DBG_TOC, ("%s: %u offsets read from sector %d\n",
775		    gp->name, nread, blk));
776		for (i = 0; i < nread; i++) {
777			sc->toc[offsets_read + i].offset =
778			    be64toh(((uint64_t *) buf)[i]);
779			sc->toc[offsets_read + i].blen = BLEN_UNDEF;
780		}
781		offsets_read += nread;
782	}
783	free(buf, M_GEOM);
784	buf = NULL;
785	offsets_read -= 1;
786	DPRINTF(GUZ_DBG_INFO, ("%s: done reading %u block offsets from %u "
787	    "sectors\n", gp->name, offsets_read, blk));
788	if (sc->nblocks != offsets_read) {
789		DPRINTF(GUZ_DBG_ERR, ("%s: read %s offsets than expected "
790		    "blocks\n", gp->name,
791		    sc->nblocks < offsets_read ? "more" : "less"));
792		goto e5;
793	}
794
795	if (type == G_UZIP) {
796		sc->dcp = g_uzip_zlib_ctor(sc->blksz);
797	} else {
798		sc->dcp = g_uzip_lzma_ctor(sc->blksz);
799	}
800	if (sc->dcp == NULL) {
801		goto e5;
802	}
803
804	/*
805	 * "Fake" last+1 block, to make it easier for the TOC parser to
806	 * iterate without making the last element a special case.
807	 */
808	sc->toc[sc->nblocks].offset = pp->mediasize;
809	/* Massage TOC (table of contents), make sure it is sound */
810	if (g_uzip_parse_toc(sc, pp, gp) != 0) {
811		DPRINTF(GUZ_DBG_ERR, ("%s: TOC error\n", gp->name));
812		goto e6;
813	}
814	mtx_init(&sc->last_mtx, "geom_uzip cache", NULL, MTX_DEF);
815	mtx_init(&sc->queue_mtx, "geom_uzip wrkthread", NULL, MTX_DEF);
816	bioq_init(&sc->bio_queue);
817	sc->last_blk = -1;
818	sc->last_buf = malloc(sc->blksz, M_GEOM_UZIP, M_WAITOK);
819	sc->req_total = 0;
820	sc->req_cached = 0;
821
822	sc->uzip_do = &g_uzip_do;
823
824	error = kproc_create(g_uzip_wrkthr, sc, &sc->procp, 0, 0, "%s",
825	    gp->name);
826	if (error != 0) {
827		goto e7;
828	}
829
830	g_topology_lock();
831	pp2 = g_new_providerf(gp, "%s", gp->name);
832	pp2->sectorsize = 512;
833	pp2->mediasize = (off_t)sc->nblocks * sc->blksz;
834	pp2->stripesize = pp->stripesize;
835	pp2->stripeoffset = pp->stripeoffset;
836	g_error_provider(pp2, 0);
837	g_access(cp, -1, 0, 0);
838
839	DPRINTF(GUZ_DBG_INFO, ("%s: taste ok (%d, %jd), (%d, %d), %x\n",
840	    gp->name, pp2->sectorsize, (intmax_t)pp2->mediasize,
841	    pp2->stripeoffset, pp2->stripesize, pp2->flags));
842	DPRINTF(GUZ_DBG_INFO, ("%s: %u x %u blocks\n", gp->name, sc->nblocks,
843	    sc->blksz));
844	return (gp);
845
846e7:
847	free(sc->last_buf, M_GEOM);
848	mtx_destroy(&sc->queue_mtx);
849	mtx_destroy(&sc->last_mtx);
850e6:
851	sc->dcp->free(sc->dcp);
852e5:
853	free(sc->toc, M_GEOM);
854e4:
855	free(gp->softc, M_GEOM_UZIP);
856e3:
857	if (buf != NULL) {
858		free(buf, M_GEOM);
859	}
860e2:
861	g_topology_lock();
862	g_access(cp, -1, 0, 0);
863e1:
864	g_detach(cp);
865	g_destroy_consumer(cp);
866	g_destroy_geom(gp);
867
868	return (NULL);
869}
870
871static int
872g_uzip_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
873{
874	struct g_provider *pp;
875
876	KASSERT(gp != NULL, ("NULL geom"));
877	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, gp->name);
878	g_topology_assert();
879
880	if (gp->softc == NULL) {
881		DPRINTF(GUZ_DBG_ERR, ("%s(%s): gp->softc == NULL\n", __func__,
882		    gp->name));
883		return (ENXIO);
884	}
885
886	pp = LIST_FIRST(&gp->provider);
887	KASSERT(pp != NULL, ("NULL provider"));
888	if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)
889		return (EBUSY);
890
891	g_wither_geom(gp, ENXIO);
892	g_uzip_softc_free(gp);
893	return (0);
894}
895
896static struct g_class g_uzip_class = {
897	.name = UZIP_CLASS_NAME,
898	.version = G_VERSION,
899	.taste = g_uzip_taste,
900	.destroy_geom = g_uzip_destroy_geom,
901
902	.start = g_uzip_start,
903	.orphan = g_uzip_orphan,
904	.access = g_uzip_access,
905	.spoiled = g_uzip_spoiled,
906};
907
908DECLARE_GEOM_CLASS(g_uzip_class, g_uzip);
909MODULE_DEPEND(g_uzip, xz, 1, 1, 1);
910MODULE_DEPEND(g_uzip, zlib, 1, 1, 1);
911MODULE_VERSION(geom_uzip, 0);
912