xref: /freebsd/sys/geom/uzip/g_uzip.c (revision 2be1a816b9ff69588e55be0a84cbe2a31efc0f2f)
1 /*-
2  * Copyright (c) 2004 Max Khon
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/endian.h>
33 #include <sys/errno.h>
34 #include <sys/kernel.h>
35 #include <sys/lock.h>
36 #include <sys/mutex.h>
37 #include <sys/malloc.h>
38 #include <sys/systm.h>
39 
40 #include <geom/geom.h>
41 #include <net/zlib.h>
42 
43 #undef GEOM_UZIP_DEBUG
44 #ifdef GEOM_UZIP_DEBUG
45 #define DPRINTF(a)	printf a
46 #else
47 #define DPRINTF(a)
48 #endif
49 
50 MALLOC_DEFINE(M_GEOM_UZIP, "geom_uzip", "GEOM UZIP data structures");
51 
52 #define UZIP_CLASS_NAME	"UZIP"
53 
54 /*
55  * Maximum allowed valid block size (to prevent foot-shooting)
56  */
57 #define MAX_BLKSZ	(MAXPHYS - MAXPHYS / 1000 - 12)
58 
59 /*
60  * Integer values (block size, number of blocks, offsets)
61  * are stored in big-endian (network) order on disk and struct cloop_header
62  * and in native order in struct g_uzip_softc
63  */
64 
65 #define CLOOP_MAGIC_LEN 128
66 static char CLOOP_MAGIC_START[] = "#!/bin/sh\n";
67 
68 struct cloop_header {
69 	char magic[CLOOP_MAGIC_LEN];	/* cloop magic */
70 	uint32_t blksz;			/* block size */
71 	uint32_t nblocks;		/* number of blocks */
72 };
73 
74 struct g_uzip_softc {
75 	uint32_t blksz;			/* block size */
76 	uint32_t nblocks;		/* number of blocks */
77 	uint64_t *offsets;
78 
79 	struct mtx last_mtx;
80 	uint32_t last_blk;		/* last blk no */
81 	char *last_buf;			/* last blk data */
82 	int req_total;			/* total requests */
83 	int req_cached;			/* cached requests */
84 };
85 
86 static void
87 g_uzip_softc_free(struct g_uzip_softc *sc, struct g_geom *gp)
88 {
89 	if (gp != NULL) {
90 		printf("%s: %d requests, %d cached\n",
91 		    gp->name, sc->req_total, sc->req_cached);
92 	}
93 	if (sc->offsets != NULL)
94 		free(sc->offsets, M_GEOM_UZIP);
95 	mtx_destroy(&sc->last_mtx);
96 	free(sc->last_buf, M_GEOM_UZIP);
97 	free(sc, M_GEOM_UZIP);
98 }
99 
100 static void *
101 z_alloc(void *nil, u_int type, u_int size)
102 {
103 	void *ptr;
104 
105 	ptr = malloc(type * size, M_GEOM_UZIP, M_NOWAIT);
106 	return ptr;
107 }
108 
109 static void
110 z_free(void *nil, void *ptr)
111 {
112 	free(ptr, M_GEOM_UZIP);
113 }
114 
115 static void
116 g_uzip_done(struct bio *bp)
117 {
118 	int err;
119 	struct bio *bp2;
120 	z_stream zs;
121 	struct g_provider *pp, *pp2;
122 	struct g_consumer *cp;
123 	struct g_geom *gp;
124 	struct g_uzip_softc *sc;
125 	off_t pos, upos;
126 	uint32_t start_blk, i;
127 	size_t bsize;
128 
129 	bp2 = bp->bio_parent;
130 	pp = bp2->bio_to;
131 	gp = pp->geom;
132 	cp = LIST_FIRST(&gp->consumer);
133 	pp2 = cp->provider;
134 	sc = gp->softc;
135 	DPRINTF(("%s: done\n", gp->name));
136 
137 	bp2->bio_error = bp->bio_error;
138 	if (bp2->bio_error != 0)
139 		goto done;
140 
141 	/*
142 	 * Uncompress data.
143 	 */
144 	zs.zalloc = z_alloc;
145 	zs.zfree = z_free;
146 	err = inflateInit(&zs);
147 	if (err != Z_OK) {
148 		bp2->bio_error = EIO;
149 		goto done;
150 	}
151 	start_blk = bp2->bio_offset / sc->blksz;
152 	bsize = pp2->sectorsize;
153 	pos = sc->offsets[start_blk] % bsize;
154 	upos = 0;
155 	DPRINTF(("%s: done: start_blk %d, pos %lld, upos %lld (%lld, %d, %d)\n",
156 	    gp->name, start_blk, pos, upos,
157 	    bp2->bio_offset, sc->blksz, bsize));
158 	for (i = start_blk; upos < bp2->bio_length; i++) {
159 		off_t len, ulen, uoff;
160 
161 		uoff = i == start_blk ? bp2->bio_offset % sc->blksz : 0;
162 		ulen = MIN(sc->blksz - uoff, bp2->bio_length - upos);
163 		len = sc->offsets[i + 1] - sc->offsets[i];
164 
165 		if (len == 0) {
166 			/* All zero block: no cache update */
167 			bzero(bp2->bio_data + upos, ulen);
168 			upos += ulen;
169 			bp2->bio_completed += ulen;
170 			continue;
171 		}
172 		zs.next_in = bp->bio_data + pos;
173 		zs.avail_in = len;
174 		zs.next_out = sc->last_buf;
175 		zs.avail_out = sc->blksz;
176 		mtx_lock(&sc->last_mtx);
177 		err = inflate(&zs, Z_FINISH);
178 		if (err != Z_STREAM_END) {
179 			sc->last_blk = -1;
180 			mtx_unlock(&sc->last_mtx);
181 			DPRINTF(("%s: done: inflate failed (%lld + %lld -> %lld + %lld + %lld)\n",
182 			    gp->name, pos, len, uoff, upos, ulen));
183 			inflateEnd(&zs);
184 			bp2->bio_error = EIO;
185 			goto done;
186 		}
187 		sc->last_blk = i;
188 		DPRINTF(("%s: done: inflated %lld + %lld -> %lld + %lld + %lld\n",
189 		    gp->name,
190 		    pos, len,
191 		    uoff, upos, ulen));
192 		memcpy(bp2->bio_data + upos, sc->last_buf + uoff, ulen);
193 		mtx_unlock(&sc->last_mtx);
194 
195 		pos += len;
196 		upos += ulen;
197 		bp2->bio_completed += ulen;
198 		err = inflateReset(&zs);
199 		if (err != Z_OK) {
200 			inflateEnd(&zs);
201 			bp2->bio_error = EIO;
202 			goto done;
203 		}
204 	}
205 	err = inflateEnd(&zs);
206 	if (err != Z_OK) {
207 		bp2->bio_error = EIO;
208 		goto done;
209 	}
210 
211 done:
212 	/*
213 	 * Finish processing the request.
214 	 */
215 	DPRINTF(("%s: done: (%d, %lld, %ld)\n",
216 	    gp->name, bp2->bio_error, bp2->bio_completed, bp2->bio_resid));
217 	free(bp->bio_data, M_GEOM_UZIP);
218 	g_destroy_bio(bp);
219 	g_io_deliver(bp2, bp2->bio_error);
220 }
221 
222 static void
223 g_uzip_start(struct bio *bp)
224 {
225 	struct bio *bp2;
226 	struct g_provider *pp, *pp2;
227 	struct g_geom *gp;
228 	struct g_consumer *cp;
229 	struct g_uzip_softc *sc;
230 	uint32_t start_blk, end_blk;
231 	size_t bsize;
232 
233 	pp = bp->bio_to;
234 	gp = pp->geom;
235 	DPRINTF(("%s: start (%d)\n", gp->name, bp->bio_cmd));
236 
237 	if (bp->bio_cmd != BIO_READ) {
238 		g_io_deliver(bp, EOPNOTSUPP);
239 		return;
240 	}
241 
242 	cp = LIST_FIRST(&gp->consumer);
243 	pp2 = cp->provider;
244 	sc = gp->softc;
245 
246 	start_blk = bp->bio_offset / sc->blksz;
247 	end_blk = (bp->bio_offset + bp->bio_length + sc->blksz - 1) / sc->blksz;
248 	KASSERT(start_blk < sc->nblocks,
249 		("start_blk out of range"));
250 	KASSERT(end_blk <= sc->nblocks,
251 		("end_blk out of range"));
252 
253 	sc->req_total++;
254 	if (start_blk + 1 == end_blk) {
255 		mtx_lock(&sc->last_mtx);
256 		if (start_blk == sc->last_blk) {
257 			off_t uoff;
258 
259 			uoff = bp->bio_offset % sc->blksz;
260 			KASSERT(bp->bio_length <= sc->blksz - uoff,
261 			    ("cached data error"));
262 			memcpy(bp->bio_data, sc->last_buf + uoff,
263 			    bp->bio_length);
264 			sc->req_cached++;
265 			mtx_unlock(&sc->last_mtx);
266 
267 			DPRINTF(("%s: start: cached 0 + %lld, %lld + 0 + %lld\n",
268 			    gp->name, bp->bio_length, uoff, bp->bio_length));
269 			bp->bio_completed = bp->bio_length;
270 			g_io_deliver(bp, 0);
271 			return;
272 		}
273 		mtx_unlock(&sc->last_mtx);
274 	}
275 
276 	bp2 = g_clone_bio(bp);
277 	if (bp2 == NULL) {
278 		g_io_deliver(bp, ENOMEM);
279 		return;
280 	}
281 	bp2->bio_done = g_uzip_done;
282 	DPRINTF(("%s: start (%d..%d), %s: %d + %lld, %s: %d + %lld\n",
283 	    gp->name, start_blk, end_blk,
284 	    pp->name, pp->sectorsize, pp->mediasize,
285 	    pp2->name, pp2->sectorsize, pp2->mediasize));
286 	bsize = pp2->sectorsize;
287 	bp2->bio_offset = sc->offsets[start_blk] - sc->offsets[start_blk] % bsize;
288 	bp2->bio_length = sc->offsets[end_blk] - bp2->bio_offset;
289 	bp2->bio_length = (bp2->bio_length + bsize - 1) / bsize * bsize;
290 	DPRINTF(("%s: start %lld + %lld -> %lld + %lld -> %lld + %lld\n",
291 	    gp->name,
292 	    bp->bio_offset, bp->bio_length,
293 	    sc->offsets[start_blk], sc->offsets[end_blk] - sc->offsets[start_blk],
294 	    bp2->bio_offset, bp2->bio_length));
295 	bp2->bio_data = malloc(bp2->bio_length, M_GEOM_UZIP, M_NOWAIT);
296 	if (bp2->bio_data == NULL) {
297 		g_destroy_bio(bp2);
298 		g_io_deliver(bp, ENOMEM);
299 		return;
300 	}
301 
302 	g_io_request(bp2, cp);
303 	DPRINTF(("%s: start ok\n", gp->name));
304 }
305 
306 static void
307 g_uzip_orphan(struct g_consumer *cp)
308 {
309 	struct g_geom *gp;
310 
311 	g_trace(G_T_TOPOLOGY, "g_uzip_orphan(%p/%s)", cp, cp->provider->name);
312 	g_topology_assert();
313 	KASSERT(cp->provider->error != 0,
314 		("g_uzip_orphan with error == 0"));
315 
316 	gp = cp->geom;
317 	g_uzip_softc_free(gp->softc, gp);
318 	gp->softc = NULL;
319 	g_wither_geom(gp, cp->provider->error);
320 }
321 
322 static int
323 g_uzip_access(struct g_provider *pp, int dr, int dw, int de)
324 {
325 	struct g_geom *gp;
326 	struct g_consumer *cp;
327 
328 	gp = pp->geom;
329 	cp = LIST_FIRST(&gp->consumer);
330 	KASSERT (cp != NULL, ("g_uzip_access but no consumer"));
331 
332 	if (cp->acw + dw > 0)
333 		return EROFS;
334 
335 	return (g_access(cp, dr, dw, de));
336 }
337 
338 static void
339 g_uzip_spoiled(struct g_consumer *cp)
340 {
341 	struct g_geom *gp;
342 
343 	gp = cp->geom;
344 	g_trace(G_T_TOPOLOGY, "g_uzip_spoiled(%p/%s)", cp, gp->name);
345 	g_topology_assert();
346 
347 	g_uzip_softc_free(gp->softc, gp);
348 	gp->softc = NULL;
349 	g_wither_geom(gp, ENXIO);
350 }
351 
352 static struct g_geom *
353 g_uzip_taste(struct g_class *mp, struct g_provider *pp, int flags)
354 {
355 	int error;
356 	uint32_t i, total_offsets, offsets_read, blk;
357 	void *buf;
358 	struct cloop_header *header;
359 	struct g_consumer *cp;
360 	struct g_geom *gp;
361 	struct g_provider *pp2;
362 	struct g_uzip_softc *sc;
363 
364 	g_trace(G_T_TOPOLOGY, "g_uzip_taste(%s,%s)", mp->name, pp->name);
365 	g_topology_assert();
366 	buf = NULL;
367 
368 	/*
369 	 * Create geom instance.
370 	 */
371 	gp = g_new_geomf(mp, "%s.uzip", pp->name);
372 	cp = g_new_consumer(gp);
373 	error = g_attach(cp, pp);
374 	if (error == 0)
375 		error = g_access(cp, 1, 0, 0);
376 	if (error) {
377 		g_detach(cp);
378 		g_destroy_consumer(cp);
379 		g_destroy_geom(gp);
380 		return (NULL);
381 	}
382 	g_topology_unlock();
383 
384 	/*
385 	 * Read cloop header, look for CLOOP magic, perform
386 	 * other validity checks.
387 	 */
388 	DPRINTF(("%s: media sectorsize %u, mediasize %lld\n",
389 	    gp->name, pp->sectorsize, pp->mediasize));
390 	buf = g_read_data(cp, 0, pp->sectorsize, NULL);
391 	if (buf == NULL)
392 		goto err;
393 	header = (struct cloop_header *) buf;
394 	if (strncmp(header->magic, CLOOP_MAGIC_START,
395 		    sizeof(CLOOP_MAGIC_START) - 1) != 0) {
396 		DPRINTF(("%s: no CLOOP magic\n", gp->name));
397 		goto err;
398 	}
399 	if (header->magic[0x0b] != 'V' || header->magic[0x0c] < '2') {
400 		DPRINTF(("%s: image version too old\n", gp->name));
401 		goto err;
402 	}
403 
404 	/*
405 	 * Initialize softc and read offsets.
406 	 */
407 	sc = malloc(sizeof(*sc), M_GEOM_UZIP, M_WAITOK | M_ZERO);
408 	gp->softc = sc;
409 	sc->blksz = ntohl(header->blksz);
410 	sc->nblocks = ntohl(header->nblocks);
411 	if (sc->blksz % 512 != 0) {
412 		printf("%s: block size (%u) should be multiple of 512.\n",
413 		    gp->name, sc->blksz);
414 		goto err;
415 	}
416 	if (sc->blksz > MAX_BLKSZ) {
417 		printf("%s: block size (%u) should not be larger than %d.\n",
418 		    gp->name, sc->blksz, MAX_BLKSZ);
419 	}
420 	total_offsets = sc->nblocks + 1;
421 	if (sizeof(struct cloop_header) +
422 	    total_offsets * sizeof(uint64_t) > pp->mediasize) {
423 		printf("%s: media too small for %u blocks\n",
424 		       gp->name, sc->nblocks);
425 		goto err;
426 	}
427 	sc->offsets = malloc(
428 	    total_offsets * sizeof(uint64_t), M_GEOM_UZIP, M_WAITOK);
429 	offsets_read = MIN(total_offsets,
430 	    (pp->sectorsize - sizeof(*header)) / sizeof(uint64_t));
431 	for (i = 0; i < offsets_read; i++)
432 		sc->offsets[i] = be64toh(((uint64_t *) (header + 1))[i]);
433 	DPRINTF(("%s: %u offsets in the first sector\n",
434 	       gp->name, offsets_read));
435 	for (blk = 1; offsets_read < total_offsets; blk++) {
436 		uint32_t nread;
437 
438 		free(buf, M_GEOM);
439 		buf = g_read_data(
440 		    cp, blk * pp->sectorsize, pp->sectorsize, NULL);
441 		if (buf == NULL)
442 			goto err;
443 		nread = MIN(total_offsets - offsets_read,
444 		     pp->sectorsize / sizeof(uint64_t));
445 		DPRINTF(("%s: %u offsets read from sector %d\n",
446 		    gp->name, nread, blk));
447 		for (i = 0; i < nread; i++) {
448 			sc->offsets[offsets_read + i] =
449 			    be64toh(((uint64_t *) buf)[i]);
450 		}
451 		offsets_read += nread;
452 	}
453 	DPRINTF(("%s: done reading offsets\n", gp->name));
454 	mtx_init(&sc->last_mtx, "geom_uzip cache", NULL, MTX_DEF);
455 	sc->last_blk = -1;
456 	sc->last_buf = malloc(sc->blksz, M_GEOM_UZIP, M_WAITOK);
457 	sc->req_total = 0;
458 	sc->req_cached = 0;
459 
460 	g_topology_lock();
461 	pp2 = g_new_providerf(gp, "%s", gp->name);
462 	pp2->sectorsize = 512;
463 	pp2->mediasize = (off_t)sc->nblocks * sc->blksz;
464         pp2->flags = pp->flags & G_PF_CANDELETE;
465         if (pp->stripesize > 0) {
466                 pp2->stripesize = pp->stripesize;
467                 pp2->stripeoffset = pp->stripeoffset;
468         }
469 	g_error_provider(pp2, 0);
470 	g_access(cp, -1, 0, 0);
471 
472 	DPRINTF(("%s: taste ok (%d, %lld), (%d, %d), %x\n",
473 	    gp->name,
474 	    pp2->sectorsize, pp2->mediasize,
475 	    pp2->stripeoffset, pp2->stripesize, pp2->flags));
476 	printf("%s: %u x %u blocks\n",
477 	       gp->name, sc->nblocks, sc->blksz);
478 	return (gp);
479 
480 err:
481 	g_topology_lock();
482 	g_access(cp, -1, 0, 0);
483 	if (buf != NULL)
484 		free(buf, M_GEOM);
485 	if (gp->softc != NULL) {
486 		g_uzip_softc_free(gp->softc, NULL);
487 		gp->softc = NULL;
488 	}
489 	g_detach(cp);
490 	g_destroy_consumer(cp);
491 	g_destroy_geom(gp);
492 	return (NULL);
493 }
494 
495 static int
496 g_uzip_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
497 {
498 	struct g_provider *pp;
499 
500 	g_trace(G_T_TOPOLOGY, "g_uzip_destroy_geom(%s, %s)", mp->name, gp->name);
501 	g_topology_assert();
502 
503 	if (gp->softc == NULL) {
504 		printf("%s(%s): gp->softc == NULL\n", __func__, gp->name);
505 		return (ENXIO);
506 	}
507 
508 	KASSERT(gp != NULL, ("NULL geom"));
509 	pp = LIST_FIRST(&gp->provider);
510 	KASSERT(pp != NULL, ("NULL provider"));
511 	if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)
512 		return (EBUSY);
513 
514 	g_uzip_softc_free(gp->softc, gp);
515 	gp->softc = NULL;
516 	g_wither_geom(gp, ENXIO);
517 	return (0);
518 }
519 
520 static struct g_class g_uzip_class = {
521 	.name = UZIP_CLASS_NAME,
522 	.version = G_VERSION,
523 	.taste = g_uzip_taste,
524 	.destroy_geom = g_uzip_destroy_geom,
525 
526 	.start = g_uzip_start,
527 	.orphan = g_uzip_orphan,
528 	.access = g_uzip_access,
529 	.spoiled = g_uzip_spoiled,
530 };
531 
532 DECLARE_GEOM_CLASS(g_uzip_class, g_uzip);
533 MODULE_DEPEND(g_uzip, zlib, 1, 1, 1);
534