xref: /freebsd/sys/geom/uzip/g_uzip.c (revision 64a0982bee3db2236df43357e70ce8dddbc21d48)
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("$FreeBSD$");
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 
56 MALLOC_DEFINE(M_GEOM_UZIP, "geom_uzip", "GEOM UZIP data structures");
57 
58 FEATURE(geom_uzip, "GEOM read-only compressed disks support");
59 
60 struct 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 
92 static char g_uzip_attach_to[MAXPATHLEN] = {"*"};
93 static char g_uzip_noattach_to[MAXPATHLEN] = {GUZ_DEV_NAME("*")};
94 TUNABLE_STR("kern.geom.uzip.attach_to", g_uzip_attach_to,
95     sizeof(g_uzip_attach_to));
96 TUNABLE_STR("kern.geom.uzip.noattach_to", g_uzip_noattach_to,
97     sizeof(g_uzip_noattach_to));
98 
99 SYSCTL_DECL(_kern_geom);
100 SYSCTL_NODE(_kern_geom, OID_AUTO, uzip, CTLFLAG_RW, 0, "GEOM_UZIP stuff");
101 static u_int g_uzip_debug = GEOM_UZIP_DBG_DEFAULT;
102 SYSCTL_UINT(_kern_geom_uzip, OID_AUTO, debug, CTLFLAG_RWTUN, &g_uzip_debug, 0,
103     "Debug level (0-4)");
104 static u_int g_uzip_debug_block = BLEN_UNDEF;
105 SYSCTL_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 
134 static char CLOOP_MAGIC_START[] = "#!/bin/sh\n";
135 
136 static void g_uzip_read_done(struct bio *bp);
137 static void g_uzip_do(struct g_uzip_softc *, struct bio *bp);
138 
139 static void
140 g_uzip_softc_free(struct g_uzip_softc *sc, struct g_geom *gp)
141 {
142 
143 	if (gp != NULL) {
144 		DPRINTF(GUZ_DBG_INFO, ("%s: %d requests, %d cached\n",
145 		    gp->name, sc->req_total, sc->req_cached));
146 	}
147 
148 	mtx_lock(&sc->queue_mtx);
149 	sc->wrkthr_flags |= GUZ_SHUTDOWN;
150 	wakeup(sc);
151 	while (!(sc->wrkthr_flags & GUZ_EXITING)) {
152 		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "guzfree",
153 		    hz / 10);
154 	}
155 	mtx_unlock(&sc->queue_mtx);
156 
157 	sc->dcp->free(sc->dcp);
158 	free(sc->toc, M_GEOM_UZIP);
159 	mtx_destroy(&sc->queue_mtx);
160 	mtx_destroy(&sc->last_mtx);
161 	free(sc->last_buf, M_GEOM_UZIP);
162 	free(sc, M_GEOM_UZIP);
163 }
164 
165 static int
166 g_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 
215 static int
216 g_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 
316 static void
317 g_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 
333 static int
334 g_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 
342 static void
343 g_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 */
389 zero_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 
440 done:
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 
450 static void
451 g_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 
478 static void
479 g_uzip_orphan(struct g_consumer *cp)
480 {
481 	struct g_geom *gp;
482 
483 	g_trace(G_T_TOPOLOGY, "%s(%p/%s)", __func__, cp, cp->provider->name);
484 	g_topology_assert();
485 
486 	gp = cp->geom;
487 	g_uzip_softc_free(gp->softc, gp);
488 	gp->softc = NULL;
489 	g_wither_geom(gp, ENXIO);
490 }
491 
492 static int
493 g_uzip_access(struct g_provider *pp, int dr, int dw, int de)
494 {
495 	struct g_geom *gp;
496 	struct g_consumer *cp;
497 
498 	gp = pp->geom;
499 	cp = LIST_FIRST(&gp->consumer);
500 	KASSERT (cp != NULL, ("g_uzip_access but no consumer"));
501 
502 	if (cp->acw + dw > 0)
503 		return (EROFS);
504 
505 	return (g_access(cp, dr, dw, de));
506 }
507 
508 static void
509 g_uzip_spoiled(struct g_consumer *cp)
510 {
511 	struct g_geom *gp;
512 
513 	G_VALID_CONSUMER(cp);
514 	gp = cp->geom;
515 	g_trace(G_T_TOPOLOGY, "%s(%p/%s)", __func__, cp, gp->name);
516 	g_topology_assert();
517 
518 	g_uzip_softc_free(gp->softc, gp);
519 	gp->softc = NULL;
520 	g_wither_geom(gp, ENXIO);
521 }
522 
523 static int
524 g_uzip_parse_toc(struct g_uzip_softc *sc, struct g_provider *pp,
525     struct g_geom *gp)
526 {
527 	uint32_t i, j, backref_to;
528 	uint64_t max_offset, min_offset;
529 	struct g_uzip_blk *last_blk;
530 
531 	min_offset = sizeof(struct cloop_header) +
532 	    (sc->nblocks + 1) * sizeof(uint64_t);
533 	max_offset = sc->toc[0].offset - 1;
534 	last_blk = &sc->toc[0];
535 	for (i = 0; i < sc->nblocks; i++) {
536 		/* First do some bounds checking */
537 		if ((sc->toc[i].offset < min_offset) ||
538 		    (sc->toc[i].offset > pp->mediasize)) {
539 			goto error_offset;
540 		}
541 		DPRINTF_BLK(GUZ_DBG_IO, i, ("%s: cluster #%u "
542 		    "offset=%ju max_offset=%ju\n", gp->name,
543 		    (u_int)i, (uintmax_t)sc->toc[i].offset,
544 		    (uintmax_t)max_offset));
545 		backref_to = BLEN_UNDEF;
546 		if (sc->toc[i].offset < max_offset) {
547 			/*
548 			 * For the backref'ed blocks search already parsed
549 			 * TOC entries for the matching offset and copy the
550 			 * size from matched entry.
551 			 */
552 			for (j = 0; j <= i; j++) {
553                                 if (sc->toc[j].offset == sc->toc[i].offset &&
554 				    !BLK_IS_NIL(sc, j)) {
555                                         break;
556                                 }
557                                 if (j != i) {
558 					continue;
559 				}
560 				DPRINTF(GUZ_DBG_ERR, ("%s: cannot match "
561 				    "backref'ed offset at cluster #%u\n",
562 				    gp->name, i));
563 				return (-1);
564 			}
565 			sc->toc[i].blen = sc->toc[j].blen;
566 			backref_to = j;
567 		} else {
568 			last_blk = &sc->toc[i];
569 			/*
570 			 * For the "normal blocks" seek forward until we hit
571 			 * block whose offset is larger than ours and assume
572 			 * it's going to be the next one.
573 			 */
574 			for (j = i + 1; j < sc->nblocks; j++) {
575 				if (sc->toc[j].offset > max_offset) {
576 					break;
577 				}
578 			}
579 			sc->toc[i].blen = sc->toc[j].offset -
580 			    sc->toc[i].offset;
581 			if (BLK_ENDS(sc, i) > pp->mediasize) {
582 				DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u "
583 				    "extends past media boundary (%ju > %ju)\n",
584 				    gp->name, (u_int)i,
585 				    (uintmax_t)BLK_ENDS(sc, i),
586 				    (intmax_t)pp->mediasize));
587 				return (-1);
588 			}
589 			KASSERT(max_offset <= sc->toc[i].offset, (
590 			    "%s: max_offset is incorrect: %ju",
591 			    gp->name, (uintmax_t)max_offset));
592 			max_offset = BLK_ENDS(sc, i) - 1;
593 		}
594 		DPRINTF_BLK(GUZ_DBG_TOC, i, ("%s: cluster #%u, original %u "
595 		    "bytes, in %u bytes", gp->name, i, sc->blksz,
596 		    sc->toc[i].blen));
597 		if (backref_to != BLEN_UNDEF) {
598 			DPRINTF_BLK(GUZ_DBG_TOC, i, (" (->#%u)",
599 			    (u_int)backref_to));
600 		}
601 		DPRINTF_BLK(GUZ_DBG_TOC, i, ("\n"));
602 	}
603 	last_blk->last = 1;
604 	/* Do a second pass to validate block lengths */
605 	for (i = 0; i < sc->nblocks; i++) {
606 		if (sc->toc[i].blen > sc->dcp->max_blen) {
607 			if (sc->toc[i].last == 0) {
608 				DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u "
609 				    "length (%ju) exceeds "
610 				    "max_blen (%ju)\n", gp->name, i,
611 				    (uintmax_t)sc->toc[i].blen,
612 				    (uintmax_t)sc->dcp->max_blen));
613 				return (-1);
614 			}
615 			DPRINTF(GUZ_DBG_INFO, ("%s: cluster #%u extra "
616 			    "padding is detected, trimmed to %ju\n",
617 			    gp->name, i, (uintmax_t)sc->dcp->max_blen));
618 			    sc->toc[i].blen = sc->dcp->max_blen;
619 			sc->toc[i].padded = 1;
620 		}
621 	}
622 	return (0);
623 
624 error_offset:
625 	DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u: invalid offset %ju, "
626 	    "min_offset=%ju mediasize=%jd\n", gp->name, (u_int)i,
627 	    sc->toc[i].offset, min_offset, pp->mediasize));
628 	return (-1);
629 }
630 
631 static struct g_geom *
632 g_uzip_taste(struct g_class *mp, struct g_provider *pp, int flags)
633 {
634 	int error;
635 	uint32_t i, total_offsets, offsets_read, blk;
636 	void *buf;
637 	struct cloop_header *header;
638 	struct g_consumer *cp;
639 	struct g_geom *gp;
640 	struct g_provider *pp2;
641 	struct g_uzip_softc *sc;
642 	enum {
643 		G_UZIP = 1,
644 		G_ULZMA
645 	} type;
646 
647 	g_trace(G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name);
648 	g_topology_assert();
649 
650 	/* Skip providers that are already open for writing. */
651 	if (pp->acw > 0)
652 		return (NULL);
653 
654 	if ((fnmatch(g_uzip_attach_to, pp->name, 0) != 0) ||
655 	    (fnmatch(g_uzip_noattach_to, pp->name, 0) == 0)) {
656 		DPRINTF(GUZ_DBG_INFO, ("%s(%s,%s), ignoring\n", __func__,
657 		    mp->name, pp->name));
658 		return (NULL);
659 	}
660 
661 	buf = NULL;
662 
663 	/*
664 	 * Create geom instance.
665 	 */
666 	gp = g_new_geomf(mp, GUZ_DEV_NAME("%s"), pp->name);
667 	cp = g_new_consumer(gp);
668 	error = g_attach(cp, pp);
669 	if (error == 0)
670 		error = g_access(cp, 1, 0, 0);
671 	if (error) {
672 		goto e1;
673 	}
674 	g_topology_unlock();
675 
676 	/*
677 	 * Read cloop header, look for CLOOP magic, perform
678 	 * other validity checks.
679 	 */
680 	DPRINTF(GUZ_DBG_INFO, ("%s: media sectorsize %u, mediasize %jd\n",
681 	    gp->name, pp->sectorsize, (intmax_t)pp->mediasize));
682 	buf = g_read_data(cp, 0, pp->sectorsize, NULL);
683 	if (buf == NULL)
684 		goto e2;
685 	header = (struct cloop_header *) buf;
686 	if (strncmp(header->magic, CLOOP_MAGIC_START,
687 	    sizeof(CLOOP_MAGIC_START) - 1) != 0) {
688 		DPRINTF(GUZ_DBG_ERR, ("%s: no CLOOP magic\n", gp->name));
689 		goto e3;
690 	}
691 
692 	switch (header->magic[CLOOP_OFS_COMPR]) {
693 	case CLOOP_COMP_LZMA:
694 	case CLOOP_COMP_LZMA_DDP:
695 		type = G_ULZMA;
696 		if (header->magic[CLOOP_OFS_VERSN] < CLOOP_MINVER_LZMA) {
697 			DPRINTF(GUZ_DBG_ERR, ("%s: image version too old\n",
698 			    gp->name));
699 			goto e3;
700 		}
701 		DPRINTF(GUZ_DBG_INFO, ("%s: GEOM_UZIP_LZMA image found\n",
702 		    gp->name));
703 		break;
704 	case CLOOP_COMP_LIBZ:
705 	case CLOOP_COMP_LIBZ_DDP:
706 		type = G_UZIP;
707 		if (header->magic[CLOOP_OFS_VERSN] < CLOOP_MINVER_ZLIB) {
708 			DPRINTF(GUZ_DBG_ERR, ("%s: image version too old\n",
709 			    gp->name));
710 			goto e3;
711 		}
712 		DPRINTF(GUZ_DBG_INFO, ("%s: GEOM_UZIP_ZLIB image found\n",
713 		    gp->name));
714 		break;
715 	default:
716 		DPRINTF(GUZ_DBG_ERR, ("%s: unsupported image type\n",
717 		    gp->name));
718                 goto e3;
719         }
720 
721 	/*
722 	 * Initialize softc and read offsets.
723 	 */
724 	sc = malloc(sizeof(*sc), M_GEOM_UZIP, M_WAITOK | M_ZERO);
725 	gp->softc = sc;
726 	sc->blksz = ntohl(header->blksz);
727 	sc->nblocks = ntohl(header->nblocks);
728 	if (sc->blksz % 512 != 0) {
729 		printf("%s: block size (%u) should be multiple of 512.\n",
730 		    gp->name, sc->blksz);
731 		goto e4;
732 	}
733 	if (sc->blksz > MAX_BLKSZ) {
734 		printf("%s: block size (%u) should not be larger than %d.\n",
735 		    gp->name, sc->blksz, MAX_BLKSZ);
736 	}
737 	total_offsets = sc->nblocks + 1;
738 	if (sizeof(struct cloop_header) +
739 	    total_offsets * sizeof(uint64_t) > pp->mediasize) {
740 		printf("%s: media too small for %u blocks\n",
741 		    gp->name, sc->nblocks);
742 		goto e4;
743 	}
744 	sc->toc = malloc(total_offsets * sizeof(struct g_uzip_blk),
745 	    M_GEOM_UZIP, M_WAITOK | M_ZERO);
746 	offsets_read = MIN(total_offsets,
747 	    (pp->sectorsize - sizeof(*header)) / sizeof(uint64_t));
748 	for (i = 0; i < offsets_read; i++) {
749 		sc->toc[i].offset = be64toh(((uint64_t *) (header + 1))[i]);
750 		sc->toc[i].blen = BLEN_UNDEF;
751 	}
752 	DPRINTF(GUZ_DBG_INFO, ("%s: %u offsets in the first sector\n",
753 	       gp->name, offsets_read));
754 	for (blk = 1; offsets_read < total_offsets; blk++) {
755 		uint32_t nread;
756 
757 		free(buf, M_GEOM);
758 		buf = g_read_data(
759 		    cp, blk * pp->sectorsize, pp->sectorsize, NULL);
760 		if (buf == NULL)
761 			goto e5;
762 		nread = MIN(total_offsets - offsets_read,
763 		     pp->sectorsize / sizeof(uint64_t));
764 		DPRINTF(GUZ_DBG_TOC, ("%s: %u offsets read from sector %d\n",
765 		    gp->name, nread, blk));
766 		for (i = 0; i < nread; i++) {
767 			sc->toc[offsets_read + i].offset =
768 			    be64toh(((uint64_t *) buf)[i]);
769 			sc->toc[offsets_read + i].blen = BLEN_UNDEF;
770 		}
771 		offsets_read += nread;
772 	}
773 	free(buf, M_GEOM);
774 	buf = NULL;
775 	offsets_read -= 1;
776 	DPRINTF(GUZ_DBG_INFO, ("%s: done reading %u block offsets from %u "
777 	    "sectors\n", gp->name, offsets_read, blk));
778 	if (sc->nblocks != offsets_read) {
779 		DPRINTF(GUZ_DBG_ERR, ("%s: read %s offsets than expected "
780 		    "blocks\n", gp->name,
781 		    sc->nblocks < offsets_read ? "more" : "less"));
782 		goto e5;
783 	}
784 
785 	if (type == G_UZIP) {
786 		sc->dcp = g_uzip_zlib_ctor(sc->blksz);
787 	} else {
788 		sc->dcp = g_uzip_lzma_ctor(sc->blksz);
789 	}
790 	if (sc->dcp == NULL) {
791 		goto e5;
792 	}
793 
794 	/*
795 	 * "Fake" last+1 block, to make it easier for the TOC parser to
796 	 * iterate without making the last element a special case.
797 	 */
798 	sc->toc[sc->nblocks].offset = pp->mediasize;
799 	/* Massage TOC (table of contents), make sure it is sound */
800 	if (g_uzip_parse_toc(sc, pp, gp) != 0) {
801 		DPRINTF(GUZ_DBG_ERR, ("%s: TOC error\n", gp->name));
802 		goto e6;
803 	}
804 	mtx_init(&sc->last_mtx, "geom_uzip cache", NULL, MTX_DEF);
805 	mtx_init(&sc->queue_mtx, "geom_uzip wrkthread", NULL, MTX_DEF);
806 	bioq_init(&sc->bio_queue);
807 	sc->last_blk = -1;
808 	sc->last_buf = malloc(sc->blksz, M_GEOM_UZIP, M_WAITOK);
809 	sc->req_total = 0;
810 	sc->req_cached = 0;
811 
812 	sc->uzip_do = &g_uzip_do;
813 
814 	error = kproc_create(g_uzip_wrkthr, sc, &sc->procp, 0, 0, "%s",
815 	    gp->name);
816 	if (error != 0) {
817 		goto e7;
818 	}
819 
820 	g_topology_lock();
821 	pp2 = g_new_providerf(gp, "%s", gp->name);
822 	pp2->sectorsize = 512;
823 	pp2->mediasize = (off_t)sc->nblocks * sc->blksz;
824 	pp2->stripesize = pp->stripesize;
825 	pp2->stripeoffset = pp->stripeoffset;
826 	g_error_provider(pp2, 0);
827 	g_access(cp, -1, 0, 0);
828 
829 	DPRINTF(GUZ_DBG_INFO, ("%s: taste ok (%d, %jd), (%d, %d), %x\n",
830 	    gp->name, pp2->sectorsize, (intmax_t)pp2->mediasize,
831 	    pp2->stripeoffset, pp2->stripesize, pp2->flags));
832 	DPRINTF(GUZ_DBG_INFO, ("%s: %u x %u blocks\n", gp->name, sc->nblocks,
833 	    sc->blksz));
834 	return (gp);
835 
836 e7:
837 	free(sc->last_buf, M_GEOM);
838 	mtx_destroy(&sc->queue_mtx);
839 	mtx_destroy(&sc->last_mtx);
840 e6:
841 	sc->dcp->free(sc->dcp);
842 e5:
843 	free(sc->toc, M_GEOM);
844 e4:
845 	free(gp->softc, M_GEOM_UZIP);
846 e3:
847 	if (buf != NULL) {
848 		free(buf, M_GEOM);
849 	}
850 e2:
851 	g_topology_lock();
852 	g_access(cp, -1, 0, 0);
853 e1:
854 	g_detach(cp);
855 	g_destroy_consumer(cp);
856 	g_destroy_geom(gp);
857 
858 	return (NULL);
859 }
860 
861 static int
862 g_uzip_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
863 {
864 	struct g_provider *pp;
865 
866 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, gp->name);
867 	g_topology_assert();
868 
869 	if (gp->softc == NULL) {
870 		DPRINTF(GUZ_DBG_ERR, ("%s(%s): gp->softc == NULL\n", __func__,
871 		    gp->name));
872 		return (ENXIO);
873 	}
874 
875 	KASSERT(gp != NULL, ("NULL geom"));
876 	pp = LIST_FIRST(&gp->provider);
877 	KASSERT(pp != NULL, ("NULL provider"));
878 	if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)
879 		return (EBUSY);
880 
881 	g_uzip_softc_free(gp->softc, gp);
882 	gp->softc = NULL;
883 	g_wither_geom(gp, ENXIO);
884 
885 	return (0);
886 }
887 
888 static struct g_class g_uzip_class = {
889 	.name = UZIP_CLASS_NAME,
890 	.version = G_VERSION,
891 	.taste = g_uzip_taste,
892 	.destroy_geom = g_uzip_destroy_geom,
893 
894 	.start = g_uzip_start,
895 	.orphan = g_uzip_orphan,
896 	.access = g_uzip_access,
897 	.spoiled = g_uzip_spoiled,
898 };
899 
900 DECLARE_GEOM_CLASS(g_uzip_class, g_uzip);
901 MODULE_DEPEND(g_uzip, zlib, 1, 1, 1);
902