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#include <sys/sysctl.h>
40
41#include <geom/geom.h>
42#include <net/zlib.h>
43
44FEATURE(geom_uzip, "GEOM uzip read-only compressed disks support");
45
46#undef GEOM_UZIP_DEBUG
47#ifdef GEOM_UZIP_DEBUG
48#define DPRINTF(a)	printf a
49#else
50#define DPRINTF(a)
51#endif
52
53static MALLOC_DEFINE(M_GEOM_UZIP, "geom_uzip", "GEOM UZIP data structures");
54
55#define UZIP_CLASS_NAME	"UZIP"
56
57/*
58 * Maximum allowed valid block size (to prevent foot-shooting)
59 */
60#define MAX_BLKSZ	(MAXPHYS - MAXPHYS / 1000 - 12)
61
62/*
63 * Integer values (block size, number of blocks, offsets)
64 * are stored in big-endian (network) order on disk and struct cloop_header
65 * and in native order in struct g_uzip_softc
66 */
67
68#define CLOOP_MAGIC_LEN 128
69static char CLOOP_MAGIC_START[] = "#!/bin/sh\n";
70
71struct cloop_header {
72	char magic[CLOOP_MAGIC_LEN];	/* cloop magic */
73	uint32_t blksz;			/* block size */
74	uint32_t nblocks;		/* number of blocks */
75};
76
77struct g_uzip_softc {
78	uint32_t blksz;			/* block size */
79	uint32_t nblocks;		/* number of blocks */
80	uint64_t *offsets;
81
82	struct mtx last_mtx;
83	uint32_t last_blk;		/* last blk no */
84	char *last_buf;			/* last blk data */
85	int req_total;			/* total requests */
86	int req_cached;			/* cached requests */
87};
88
89static void
90g_uzip_softc_free(struct g_uzip_softc *sc, struct g_geom *gp)
91{
92	if (gp != NULL) {
93		printf("%s: %d requests, %d cached\n",
94		    gp->name, sc->req_total, sc->req_cached);
95	}
96	if (sc->offsets != NULL)
97		free(sc->offsets, M_GEOM_UZIP);
98	mtx_destroy(&sc->last_mtx);
99	free(sc->last_buf, M_GEOM_UZIP);
100	free(sc, M_GEOM_UZIP);
101}
102
103static void *
104z_alloc(void *nil, u_int type, u_int size)
105{
106	void *ptr;
107
108	ptr = malloc(type * size, M_GEOM_UZIP, M_NOWAIT);
109	return ptr;
110}
111
112static void
113z_free(void *nil, void *ptr)
114{
115	free(ptr, M_GEOM_UZIP);
116}
117
118static void
119g_uzip_done(struct bio *bp)
120{
121	int err;
122	struct bio *bp2;
123	z_stream zs;
124	struct g_provider *pp, *pp2;
125	struct g_consumer *cp;
126	struct g_geom *gp;
127	struct g_uzip_softc *sc;
128	off_t pos, upos;
129	uint32_t start_blk, i;
130	size_t bsize;
131
132	bp2 = bp->bio_parent;
133	pp = bp2->bio_to;
134	gp = pp->geom;
135	cp = LIST_FIRST(&gp->consumer);
136	pp2 = cp->provider;
137	sc = gp->softc;
138	DPRINTF(("%s: done\n", gp->name));
139
140	bp2->bio_error = bp->bio_error;
141	if (bp2->bio_error != 0)
142		goto done;
143
144	/*
145	 * Uncompress data.
146	 */
147	zs.zalloc = z_alloc;
148	zs.zfree = z_free;
149	err = inflateInit(&zs);
150	if (err != Z_OK) {
151		bp2->bio_error = EIO;
152		goto done;
153	}
154	start_blk = bp2->bio_offset / sc->blksz;
155	bsize = pp2->sectorsize;
156	pos = sc->offsets[start_blk] % bsize;
157	upos = 0;
158	DPRINTF(("%s: done: start_blk %d, pos %lld, upos %lld (%lld, %d, %d)\n",
159	    gp->name, start_blk, pos, upos,
160	    bp2->bio_offset, sc->blksz, bsize));
161	for (i = start_blk; upos < bp2->bio_length; i++) {
162		off_t len, ulen, uoff;
163
164		uoff = i == start_blk ? bp2->bio_offset % sc->blksz : 0;
165		ulen = MIN(sc->blksz - uoff, bp2->bio_length - upos);
166		len = sc->offsets[i + 1] - sc->offsets[i];
167
168		if (len == 0) {
169			/* All zero block: no cache update */
170			bzero(bp2->bio_data + upos, ulen);
171			upos += ulen;
172			bp2->bio_completed += ulen;
173			continue;
174		}
175		zs.next_in = bp->bio_data + pos;
176		zs.avail_in = len;
177		zs.next_out = sc->last_buf;
178		zs.avail_out = sc->blksz;
179		mtx_lock(&sc->last_mtx);
180		err = inflate(&zs, Z_FINISH);
181		if (err != Z_STREAM_END) {
182			sc->last_blk = -1;
183			mtx_unlock(&sc->last_mtx);
184			DPRINTF(("%s: done: inflate failed (%lld + %lld -> %lld + %lld + %lld)\n",
185			    gp->name, pos, len, uoff, upos, ulen));
186			inflateEnd(&zs);
187			bp2->bio_error = EIO;
188			goto done;
189		}
190		sc->last_blk = i;
191		DPRINTF(("%s: done: inflated %lld + %lld -> %lld + %lld + %lld\n",
192		    gp->name,
193		    pos, len,
194		    uoff, upos, ulen));
195		memcpy(bp2->bio_data + upos, sc->last_buf + uoff, ulen);
196		mtx_unlock(&sc->last_mtx);
197
198		pos += len;
199		upos += ulen;
200		bp2->bio_completed += ulen;
201		err = inflateReset(&zs);
202		if (err != Z_OK) {
203			inflateEnd(&zs);
204			bp2->bio_error = EIO;
205			goto done;
206		}
207	}
208	err = inflateEnd(&zs);
209	if (err != Z_OK) {
210		bp2->bio_error = EIO;
211		goto done;
212	}
213
214done:
215	/*
216	 * Finish processing the request.
217	 */
218	DPRINTF(("%s: done: (%d, %lld, %ld)\n",
219	    gp->name, bp2->bio_error, bp2->bio_completed, bp2->bio_resid));
220	free(bp->bio_data, M_GEOM_UZIP);
221	g_destroy_bio(bp);
222	g_io_deliver(bp2, bp2->bio_error);
223}
224
225static void
226g_uzip_start(struct bio *bp)
227{
228	struct bio *bp2;
229	struct g_provider *pp, *pp2;
230	struct g_geom *gp;
231	struct g_consumer *cp;
232	struct g_uzip_softc *sc;
233	uint32_t start_blk, end_blk;
234	size_t bsize;
235
236	pp = bp->bio_to;
237	gp = pp->geom;
238	DPRINTF(("%s: start (%d)\n", gp->name, bp->bio_cmd));
239
240	if (bp->bio_cmd != BIO_READ) {
241		g_io_deliver(bp, EOPNOTSUPP);
242		return;
243	}
244
245	cp = LIST_FIRST(&gp->consumer);
246	pp2 = cp->provider;
247	sc = gp->softc;
248
249	start_blk = bp->bio_offset / sc->blksz;
250	end_blk = (bp->bio_offset + bp->bio_length + sc->blksz - 1) / sc->blksz;
251	KASSERT(start_blk < sc->nblocks,
252		("start_blk out of range"));
253	KASSERT(end_blk <= sc->nblocks,
254		("end_blk out of range"));
255
256	sc->req_total++;
257	if (start_blk + 1 == end_blk) {
258		mtx_lock(&sc->last_mtx);
259		if (start_blk == sc->last_blk) {
260			off_t uoff;
261
262			uoff = bp->bio_offset % sc->blksz;
263			KASSERT(bp->bio_length <= sc->blksz - uoff,
264			    ("cached data error"));
265			memcpy(bp->bio_data, sc->last_buf + uoff,
266			    bp->bio_length);
267			sc->req_cached++;
268			mtx_unlock(&sc->last_mtx);
269
270			DPRINTF(("%s: start: cached 0 + %lld, %lld + 0 + %lld\n",
271			    gp->name, bp->bio_length, uoff, bp->bio_length));
272			bp->bio_completed = bp->bio_length;
273			g_io_deliver(bp, 0);
274			return;
275		}
276		mtx_unlock(&sc->last_mtx);
277	}
278
279	bp2 = g_clone_bio(bp);
280	if (bp2 == NULL) {
281		g_io_deliver(bp, ENOMEM);
282		return;
283	}
284	bp2->bio_done = g_uzip_done;
285	DPRINTF(("%s: start (%d..%d), %s: %d + %lld, %s: %d + %lld\n",
286	    gp->name, start_blk, end_blk,
287	    pp->name, pp->sectorsize, pp->mediasize,
288	    pp2->name, pp2->sectorsize, pp2->mediasize));
289	bsize = pp2->sectorsize;
290	bp2->bio_offset = sc->offsets[start_blk] - sc->offsets[start_blk] % bsize;
291	bp2->bio_length = sc->offsets[end_blk] - bp2->bio_offset;
292	bp2->bio_length = (bp2->bio_length + bsize - 1) / bsize * bsize;
293	DPRINTF(("%s: start %lld + %lld -> %lld + %lld -> %lld + %lld\n",
294	    gp->name,
295	    bp->bio_offset, bp->bio_length,
296	    sc->offsets[start_blk], sc->offsets[end_blk] - sc->offsets[start_blk],
297	    bp2->bio_offset, bp2->bio_length));
298	bp2->bio_data = malloc(bp2->bio_length, M_GEOM_UZIP, M_NOWAIT);
299	if (bp2->bio_data == NULL) {
300		g_destroy_bio(bp2);
301		g_io_deliver(bp, ENOMEM);
302		return;
303	}
304
305	g_io_request(bp2, cp);
306	DPRINTF(("%s: start ok\n", gp->name));
307}
308
309static void
310g_uzip_orphan(struct g_consumer *cp)
311{
312	struct g_geom *gp;
313
314	g_trace(G_T_TOPOLOGY, "g_uzip_orphan(%p/%s)", cp, cp->provider->name);
315	g_topology_assert();
316
317	gp = cp->geom;
318	g_uzip_softc_free(gp->softc, gp);
319	gp->softc = NULL;
320	g_wither_geom(gp, ENXIO);
321}
322
323static int
324g_uzip_access(struct g_provider *pp, int dr, int dw, int de)
325{
326	struct g_geom *gp;
327	struct g_consumer *cp;
328
329	gp = pp->geom;
330	cp = LIST_FIRST(&gp->consumer);
331	KASSERT (cp != NULL, ("g_uzip_access but no consumer"));
332
333	if (cp->acw + dw > 0)
334		return EROFS;
335
336	return (g_access(cp, dr, dw, de));
337}
338
339static void
340g_uzip_spoiled(struct g_consumer *cp)
341{
342	struct g_geom *gp;
343
344	gp = cp->geom;
345	g_trace(G_T_TOPOLOGY, "g_uzip_spoiled(%p/%s)", cp, gp->name);
346	g_topology_assert();
347
348	g_uzip_softc_free(gp->softc, gp);
349	gp->softc = NULL;
350	g_wither_geom(gp, ENXIO);
351}
352
353static struct g_geom *
354g_uzip_taste(struct g_class *mp, struct g_provider *pp, int flags)
355{
356	int error;
357	uint32_t i, total_offsets, offsets_read, blk;
358	void *buf;
359	struct cloop_header *header;
360	struct g_consumer *cp;
361	struct g_geom *gp;
362	struct g_provider *pp2;
363	struct g_uzip_softc *sc;
364
365	g_trace(G_T_TOPOLOGY, "g_uzip_taste(%s,%s)", mp->name, pp->name);
366	g_topology_assert();
367
368	/* Skip providers that are already open for writing. */
369	if (pp->acw > 0)
370		return (NULL);
371
372	buf = NULL;
373
374	/*
375	 * Create geom instance.
376	 */
377	gp = g_new_geomf(mp, "%s.uzip", pp->name);
378	cp = g_new_consumer(gp);
379	error = g_attach(cp, pp);
380	if (error == 0)
381		error = g_access(cp, 1, 0, 0);
382	if (error) {
383		g_detach(cp);
384		g_destroy_consumer(cp);
385		g_destroy_geom(gp);
386		return (NULL);
387	}
388	g_topology_unlock();
389
390	/*
391	 * Read cloop header, look for CLOOP magic, perform
392	 * other validity checks.
393	 */
394	DPRINTF(("%s: media sectorsize %u, mediasize %lld\n",
395	    gp->name, pp->sectorsize, pp->mediasize));
396	buf = g_read_data(cp, 0, pp->sectorsize, NULL);
397	if (buf == NULL)
398		goto err;
399	header = (struct cloop_header *) buf;
400	if (strncmp(header->magic, CLOOP_MAGIC_START,
401		    sizeof(CLOOP_MAGIC_START) - 1) != 0) {
402		DPRINTF(("%s: no CLOOP magic\n", gp->name));
403		goto err;
404	}
405	if (header->magic[0x0b] != 'V' || header->magic[0x0c] < '2') {
406		DPRINTF(("%s: image version too old\n", gp->name));
407		goto err;
408	}
409
410	/*
411	 * Initialize softc and read offsets.
412	 */
413	sc = malloc(sizeof(*sc), M_GEOM_UZIP, M_WAITOK | M_ZERO);
414	gp->softc = sc;
415	sc->blksz = ntohl(header->blksz);
416	sc->nblocks = ntohl(header->nblocks);
417	if (sc->blksz % 512 != 0) {
418		printf("%s: block size (%u) should be multiple of 512.\n",
419		    gp->name, sc->blksz);
420		goto err;
421	}
422	if (sc->blksz > MAX_BLKSZ) {
423		printf("%s: block size (%u) should not be larger than %d.\n",
424		    gp->name, sc->blksz, MAX_BLKSZ);
425	}
426	total_offsets = sc->nblocks + 1;
427	if (sizeof(struct cloop_header) +
428	    total_offsets * sizeof(uint64_t) > pp->mediasize) {
429		printf("%s: media too small for %u blocks\n",
430		       gp->name, sc->nblocks);
431		goto err;
432	}
433	sc->offsets = malloc(
434	    total_offsets * sizeof(uint64_t), M_GEOM_UZIP, M_WAITOK);
435	offsets_read = MIN(total_offsets,
436	    (pp->sectorsize - sizeof(*header)) / sizeof(uint64_t));
437	for (i = 0; i < offsets_read; i++)
438		sc->offsets[i] = be64toh(((uint64_t *) (header + 1))[i]);
439	DPRINTF(("%s: %u offsets in the first sector\n",
440	       gp->name, offsets_read));
441	for (blk = 1; offsets_read < total_offsets; blk++) {
442		uint32_t nread;
443
444		free(buf, M_GEOM);
445		buf = g_read_data(
446		    cp, blk * pp->sectorsize, pp->sectorsize, NULL);
447		if (buf == NULL)
448			goto err;
449		nread = MIN(total_offsets - offsets_read,
450		     pp->sectorsize / sizeof(uint64_t));
451		DPRINTF(("%s: %u offsets read from sector %d\n",
452		    gp->name, nread, blk));
453		for (i = 0; i < nread; i++) {
454			sc->offsets[offsets_read + i] =
455			    be64toh(((uint64_t *) buf)[i]);
456		}
457		offsets_read += nread;
458	}
459	DPRINTF(("%s: done reading offsets\n", gp->name));
460	mtx_init(&sc->last_mtx, "geom_uzip cache", NULL, MTX_DEF);
461	sc->last_blk = -1;
462	sc->last_buf = malloc(sc->blksz, M_GEOM_UZIP, M_WAITOK);
463	sc->req_total = 0;
464	sc->req_cached = 0;
465
466	g_topology_lock();
467	pp2 = g_new_providerf(gp, "%s", gp->name);
468	pp2->sectorsize = 512;
469	pp2->mediasize = (off_t)sc->nblocks * sc->blksz;
470        pp2->flags = pp->flags & G_PF_CANDELETE;
471        pp2->stripesize = pp->stripesize;
472        pp2->stripeoffset = pp->stripeoffset;
473	g_error_provider(pp2, 0);
474	g_access(cp, -1, 0, 0);
475
476	DPRINTF(("%s: taste ok (%d, %lld), (%d, %d), %x\n",
477	    gp->name,
478	    pp2->sectorsize, pp2->mediasize,
479	    pp2->stripeoffset, pp2->stripesize, pp2->flags));
480	printf("%s: %u x %u blocks\n",
481	       gp->name, sc->nblocks, sc->blksz);
482	return (gp);
483
484err:
485	g_topology_lock();
486	g_access(cp, -1, 0, 0);
487	if (buf != NULL)
488		free(buf, M_GEOM);
489	if (gp->softc != NULL) {
490		g_uzip_softc_free(gp->softc, NULL);
491		gp->softc = NULL;
492	}
493	g_detach(cp);
494	g_destroy_consumer(cp);
495	g_destroy_geom(gp);
496	return (NULL);
497}
498
499static int
500g_uzip_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
501{
502	struct g_provider *pp;
503
504	g_trace(G_T_TOPOLOGY, "g_uzip_destroy_geom(%s, %s)", mp->name, gp->name);
505	g_topology_assert();
506
507	if (gp->softc == NULL) {
508		printf("%s(%s): gp->softc == NULL\n", __func__, gp->name);
509		return (ENXIO);
510	}
511
512	KASSERT(gp != NULL, ("NULL geom"));
513	pp = LIST_FIRST(&gp->provider);
514	KASSERT(pp != NULL, ("NULL provider"));
515	if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)
516		return (EBUSY);
517
518	g_uzip_softc_free(gp->softc, gp);
519	gp->softc = NULL;
520	g_wither_geom(gp, ENXIO);
521	return (0);
522}
523
524static struct g_class g_uzip_class = {
525	.name = UZIP_CLASS_NAME,
526	.version = G_VERSION,
527	.taste = g_uzip_taste,
528	.destroy_geom = g_uzip_destroy_geom,
529
530	.start = g_uzip_start,
531	.orphan = g_uzip_orphan,
532	.access = g_uzip_access,
533	.spoiled = g_uzip_spoiled,
534};
535
536DECLARE_GEOM_CLASS(g_uzip_class, g_uzip);
537MODULE_DEPEND(g_uzip, zlib, 1, 1, 1);
538