1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2004 Max Khon
5 * Copyright (c) 2014 Juniper Networks, Inc.
6 * Copyright (c) 2006-2016 Maxim Sobolev <sobomax@FreeBSD.org>
7 * All rights reserved.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31 #include <sys/cdefs.h>
32 #include "opt_geom.h"
33 #include "opt_zstdio.h"
34
35 #include <sys/param.h>
36 #include <sys/bio.h>
37 #include <sys/endian.h>
38 #include <sys/errno.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/malloc.h>
43 #include <sys/sysctl.h>
44 #include <sys/systm.h>
45 #include <sys/kthread.h>
46
47 #include <geom/geom.h>
48
49 #include <geom/uzip/g_uzip.h>
50 #include <geom/uzip/g_uzip_cloop.h>
51 #include <geom/uzip/g_uzip_softc.h>
52 #include <geom/uzip/g_uzip_dapi.h>
53 #include <geom/uzip/g_uzip_zlib.h>
54 #include <geom/uzip/g_uzip_lzma.h>
55 #ifdef ZSTDIO
56 #include <geom/uzip/g_uzip_zstd.h>
57 #endif
58 #include <geom/uzip/g_uzip_wrkthr.h>
59
60 MALLOC_DEFINE(M_GEOM_UZIP, "geom_uzip", "GEOM UZIP data structures");
61
62 FEATURE(geom_uzip, "GEOM read-only compressed disks support");
63
64 struct g_uzip_blk {
65 uint64_t offset;
66 uint32_t blen;
67 unsigned char last:1;
68 unsigned char padded:1;
69 #define BLEN_UNDEF UINT32_MAX
70 };
71
72 #ifndef ABS
73 #define ABS(a) ((a) < 0 ? -(a) : (a))
74 #endif
75
76 #define BLK_IN_RANGE(mcn, bcn, ilen) \
77 (((bcn) != BLEN_UNDEF) && ( \
78 ((ilen) >= 0 && (mcn >= bcn) && (mcn <= ((intmax_t)(bcn) + (ilen)))) || \
79 ((ilen) < 0 && (mcn <= bcn) && (mcn >= ((intmax_t)(bcn) + (ilen)))) \
80 ))
81
82 #ifdef GEOM_UZIP_DEBUG
83 # define GEOM_UZIP_DBG_DEFAULT 3
84 #else
85 # define GEOM_UZIP_DBG_DEFAULT 0
86 #endif
87
88 #define GUZ_DBG_ERR 1
89 #define GUZ_DBG_INFO 2
90 #define GUZ_DBG_IO 3
91 #define GUZ_DBG_TOC 4
92
93 #define GUZ_DEV_SUFX ".uzip"
94 #define GUZ_DEV_NAME(p) (p GUZ_DEV_SUFX)
95
96 static char g_uzip_attach_to[MAXPATHLEN] = {"*"};
97 static char g_uzip_noattach_to[MAXPATHLEN] = {GUZ_DEV_NAME("*")};
98 TUNABLE_STR("kern.geom.uzip.attach_to", g_uzip_attach_to,
99 sizeof(g_uzip_attach_to));
100 TUNABLE_STR("kern.geom.uzip.noattach_to", g_uzip_noattach_to,
101 sizeof(g_uzip_noattach_to));
102
103 SYSCTL_DECL(_kern_geom);
104 SYSCTL_NODE(_kern_geom, OID_AUTO, uzip, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
105 "GEOM_UZIP stuff");
106 static u_int g_uzip_debug = GEOM_UZIP_DBG_DEFAULT;
107 SYSCTL_UINT(_kern_geom_uzip, OID_AUTO, debug, CTLFLAG_RWTUN, &g_uzip_debug, 0,
108 "Debug level (0-4)");
109 static u_int g_uzip_debug_block = BLEN_UNDEF;
110 SYSCTL_UINT(_kern_geom_uzip, OID_AUTO, debug_block, CTLFLAG_RWTUN,
111 &g_uzip_debug_block, 0, "Debug operations around specific cluster#");
112
113 #define DPRINTF(lvl, a) \
114 if ((lvl) <= g_uzip_debug) { \
115 printf a; \
116 }
117 #define DPRINTF_BLK(lvl, cn, a) \
118 if ((lvl) <= g_uzip_debug || \
119 BLK_IN_RANGE(cn, g_uzip_debug_block, 8) || \
120 BLK_IN_RANGE(cn, g_uzip_debug_block, -8)) { \
121 printf a; \
122 }
123 #define DPRINTF_BRNG(lvl, bcn, ecn, a) \
124 KASSERT(bcn < ecn, ("DPRINTF_BRNG: invalid range (%ju, %ju)", \
125 (uintmax_t)bcn, (uintmax_t)ecn)); \
126 if (((lvl) <= g_uzip_debug) || \
127 BLK_IN_RANGE(g_uzip_debug_block, bcn, \
128 (intmax_t)ecn - (intmax_t)bcn)) { \
129 printf a; \
130 }
131
132 #define UZIP_CLASS_NAME "UZIP"
133
134 /*
135 * Maximum allowed valid block size (to prevent foot-shooting)
136 */
137 #define MAX_BLKSZ (maxphys)
138
139 static char CLOOP_MAGIC_START[] = "#!/bin/sh\n";
140
141 static void g_uzip_read_done(struct bio *bp);
142 static void g_uzip_do(struct g_uzip_softc *, struct bio *bp);
143
144 static void
g_uzip_softc_free(struct g_geom * gp)145 g_uzip_softc_free(struct g_geom *gp)
146 {
147 struct g_uzip_softc *sc = gp->softc;
148
149 DPRINTF(GUZ_DBG_INFO, ("%s: %d requests, %d cached\n",
150 gp->name, sc->req_total, sc->req_cached));
151
152 mtx_lock(&sc->queue_mtx);
153 sc->wrkthr_flags |= GUZ_SHUTDOWN;
154 wakeup(sc);
155 while (!(sc->wrkthr_flags & GUZ_EXITING)) {
156 msleep(sc->procp, &sc->queue_mtx, PRIBIO, "guzfree",
157 hz / 10);
158 }
159 mtx_unlock(&sc->queue_mtx);
160
161 sc->dcp->free(sc->dcp);
162 free(sc->toc, M_GEOM_UZIP);
163 mtx_destroy(&sc->queue_mtx);
164 mtx_destroy(&sc->last_mtx);
165 free(sc->last_buf, M_GEOM_UZIP);
166 free(sc, M_GEOM_UZIP);
167 gp->softc = NULL;
168 }
169
170 static int
g_uzip_cached(struct g_geom * gp,struct bio * bp)171 g_uzip_cached(struct g_geom *gp, struct bio *bp)
172 {
173 struct g_uzip_softc *sc;
174 off_t ofs;
175 size_t blk, blkofs, usz;
176
177 sc = gp->softc;
178 ofs = bp->bio_offset + bp->bio_completed;
179 blk = ofs / sc->blksz;
180 mtx_lock(&sc->last_mtx);
181 if (blk == sc->last_blk) {
182 blkofs = ofs % sc->blksz;
183 usz = sc->blksz - blkofs;
184 if (bp->bio_resid < usz)
185 usz = bp->bio_resid;
186 memcpy(bp->bio_data + bp->bio_completed, sc->last_buf + blkofs,
187 usz);
188 sc->req_cached++;
189 mtx_unlock(&sc->last_mtx);
190
191 DPRINTF(GUZ_DBG_IO, ("%s/%s: %p: offset=%jd: got %jd bytes "
192 "from cache\n", __func__, gp->name, bp, (intmax_t)ofs,
193 (intmax_t)usz));
194
195 bp->bio_completed += usz;
196 bp->bio_resid -= usz;
197
198 if (bp->bio_resid == 0) {
199 g_io_deliver(bp, 0);
200 return (1);
201 }
202 } else
203 mtx_unlock(&sc->last_mtx);
204
205 return (0);
206 }
207
208 #define BLK_ENDS(sc, bi) ((sc)->toc[(bi)].offset + \
209 (sc)->toc[(bi)].blen)
210
211 #define BLK_IS_CONT(sc, bi) (BLK_ENDS((sc), (bi) - 1) == \
212 (sc)->toc[(bi)].offset)
213 #define BLK_IS_NIL(sc, bi) ((sc)->toc[(bi)].blen == 0)
214
215 #define TOFF_2_BOFF(sc, pp, bi) ((sc)->toc[(bi)].offset - \
216 (sc)->toc[(bi)].offset % (pp)->sectorsize)
217 #define TLEN_2_BLEN(sc, pp, bp, ei) roundup(BLK_ENDS((sc), (ei)) - \
218 (bp)->bio_offset, (pp)->sectorsize)
219
220 static int
g_uzip_request(struct g_geom * gp,struct bio * bp)221 g_uzip_request(struct g_geom *gp, struct bio *bp)
222 {
223 struct g_uzip_softc *sc;
224 struct bio *bp2;
225 struct g_consumer *cp;
226 struct g_provider *pp;
227 off_t ofs, start_blk_ofs;
228 size_t i, start_blk, end_blk, zsize;
229
230 if (g_uzip_cached(gp, bp) != 0)
231 return (1);
232
233 sc = gp->softc;
234
235 cp = LIST_FIRST(&gp->consumer);
236 pp = cp->provider;
237
238 ofs = bp->bio_offset + bp->bio_completed;
239 start_blk = ofs / sc->blksz;
240 KASSERT(start_blk < sc->nblocks, ("start_blk out of range"));
241 end_blk = howmany(ofs + bp->bio_resid, sc->blksz);
242 KASSERT(end_blk <= sc->nblocks, ("end_blk out of range"));
243
244 for (; BLK_IS_NIL(sc, start_blk) && start_blk < end_blk; start_blk++) {
245 /* Fill in any leading Nil blocks */
246 start_blk_ofs = ofs % sc->blksz;
247 zsize = MIN(sc->blksz - start_blk_ofs, bp->bio_resid);
248 DPRINTF_BLK(GUZ_DBG_IO, start_blk, ("%s/%s: %p/%ju: "
249 "filling %ju zero bytes\n", __func__, gp->name, gp,
250 (uintmax_t)bp->bio_completed, (uintmax_t)zsize));
251 bzero(bp->bio_data + bp->bio_completed, zsize);
252 bp->bio_completed += zsize;
253 bp->bio_resid -= zsize;
254 ofs += zsize;
255 }
256
257 if (start_blk == end_blk) {
258 KASSERT(bp->bio_resid == 0, ("bp->bio_resid is invalid"));
259 /*
260 * No non-Nil data is left, complete request immediately.
261 */
262 DPRINTF(GUZ_DBG_IO, ("%s/%s: %p: all done returning %ju "
263 "bytes\n", __func__, gp->name, gp,
264 (uintmax_t)bp->bio_completed));
265 g_io_deliver(bp, 0);
266 return (1);
267 }
268
269 for (i = start_blk + 1; i < end_blk; i++) {
270 /* Trim discontinuous areas if any */
271 if (!BLK_IS_CONT(sc, i)) {
272 end_blk = i;
273 break;
274 }
275 }
276
277 DPRINTF_BRNG(GUZ_DBG_IO, start_blk, end_blk, ("%s/%s: %p: "
278 "start=%u (%ju[%jd]), end=%u (%ju)\n", __func__, gp->name, bp,
279 (u_int)start_blk, (uintmax_t)sc->toc[start_blk].offset,
280 (intmax_t)sc->toc[start_blk].blen,
281 (u_int)end_blk, (uintmax_t)BLK_ENDS(sc, end_blk - 1)));
282
283 bp2 = g_clone_bio(bp);
284 if (bp2 == NULL) {
285 g_io_deliver(bp, ENOMEM);
286 return (1);
287 }
288 bp2->bio_done = g_uzip_read_done;
289
290 bp2->bio_offset = TOFF_2_BOFF(sc, pp, start_blk);
291 while (1) {
292 bp2->bio_length = TLEN_2_BLEN(sc, pp, bp2, end_blk - 1);
293 if (bp2->bio_length <= maxphys) {
294 break;
295 }
296 if (end_blk == (start_blk + 1)) {
297 break;
298 }
299 end_blk--;
300 }
301
302 DPRINTF(GUZ_DBG_IO, ("%s/%s: bp2->bio_length = %jd, "
303 "bp2->bio_offset = %jd\n", __func__, gp->name,
304 (intmax_t)bp2->bio_length, (intmax_t)bp2->bio_offset));
305
306 bp2->bio_data = malloc(bp2->bio_length, M_GEOM_UZIP, M_NOWAIT);
307 if (bp2->bio_data == NULL) {
308 g_destroy_bio(bp2);
309 g_io_deliver(bp, ENOMEM);
310 return (1);
311 }
312
313 DPRINTF_BRNG(GUZ_DBG_IO, start_blk, end_blk, ("%s/%s: %p: "
314 "reading %jd bytes from offset %jd\n", __func__, gp->name, bp,
315 (intmax_t)bp2->bio_length, (intmax_t)bp2->bio_offset));
316
317 g_io_request(bp2, cp);
318 return (0);
319 }
320
321 static void
g_uzip_read_done(struct bio * bp)322 g_uzip_read_done(struct bio *bp)
323 {
324 struct bio *bp2;
325 struct g_geom *gp;
326 struct g_uzip_softc *sc;
327
328 bp2 = bp->bio_parent;
329 gp = bp2->bio_to->geom;
330 sc = gp->softc;
331
332 mtx_lock(&sc->queue_mtx);
333 bioq_disksort(&sc->bio_queue, bp);
334 mtx_unlock(&sc->queue_mtx);
335 wakeup(sc);
336 }
337
338 static int
g_uzip_memvcmp(const void * memory,unsigned char val,size_t size)339 g_uzip_memvcmp(const void *memory, unsigned char val, size_t size)
340 {
341 const u_char *mm;
342
343 mm = (const u_char *)memory;
344 return (*mm == val) && memcmp(mm, mm + 1, size - 1) == 0;
345 }
346
347 static void
g_uzip_do(struct g_uzip_softc * sc,struct bio * bp)348 g_uzip_do(struct g_uzip_softc *sc, struct bio *bp)
349 {
350 struct bio *bp2;
351 struct g_provider *pp;
352 struct g_consumer *cp;
353 struct g_geom *gp;
354 char *data, *data2;
355 off_t ofs;
356 size_t blk, blkofs, len, ulen, firstblk;
357 int err;
358
359 bp2 = bp->bio_parent;
360 gp = bp2->bio_to->geom;
361
362 cp = LIST_FIRST(&gp->consumer);
363 pp = cp->provider;
364
365 bp2->bio_error = bp->bio_error;
366 if (bp2->bio_error != 0)
367 goto done;
368
369 /* Make sure there's forward progress. */
370 if (bp->bio_completed == 0) {
371 bp2->bio_error = ECANCELED;
372 goto done;
373 }
374
375 ofs = bp2->bio_offset + bp2->bio_completed;
376 firstblk = blk = ofs / sc->blksz;
377 blkofs = ofs % sc->blksz;
378 data = bp->bio_data + sc->toc[blk].offset % pp->sectorsize;
379 data2 = bp2->bio_data + bp2->bio_completed;
380 while (bp->bio_completed && bp2->bio_resid) {
381 if (blk > firstblk && !BLK_IS_CONT(sc, blk)) {
382 DPRINTF_BLK(GUZ_DBG_IO, blk, ("%s/%s: %p: backref'ed "
383 "cluster #%u requested, looping around\n",
384 __func__, gp->name, bp2, (u_int)blk));
385 goto done;
386 }
387 ulen = MIN(sc->blksz - blkofs, bp2->bio_resid);
388 len = sc->toc[blk].blen;
389 DPRINTF(GUZ_DBG_IO, ("%s/%s: %p/%ju: data2=%p, ulen=%u, "
390 "data=%p, len=%u\n", __func__, gp->name, gp,
391 bp->bio_completed, data2, (u_int)ulen, data, (u_int)len));
392 if (len == 0) {
393 /* All zero block: no cache update */
394 zero_block:
395 bzero(data2, ulen);
396 } else if (len <= bp->bio_completed) {
397 mtx_lock(&sc->last_mtx);
398 err = sc->dcp->decompress(sc->dcp, gp->name, data,
399 len, sc->last_buf);
400 if (err != 0 && sc->toc[blk].last != 0) {
401 /*
402 * Last block decompression has failed, check
403 * if it's just zero padding.
404 */
405 if (g_uzip_memvcmp(data, '\0', len) == 0) {
406 sc->toc[blk].blen = 0;
407 sc->last_blk = -1;
408 mtx_unlock(&sc->last_mtx);
409 len = 0;
410 goto zero_block;
411 }
412 }
413 if (err != 0) {
414 sc->last_blk = -1;
415 mtx_unlock(&sc->last_mtx);
416 bp2->bio_error = EILSEQ;
417 DPRINTF(GUZ_DBG_ERR, ("%s/%s: decompress"
418 "(%p, %ju, %ju) failed\n", __func__,
419 gp->name, sc->dcp, (uintmax_t)blk,
420 (uintmax_t)len));
421 goto done;
422 }
423 sc->last_blk = blk;
424 memcpy(data2, sc->last_buf + blkofs, ulen);
425 mtx_unlock(&sc->last_mtx);
426 err = sc->dcp->rewind(sc->dcp, gp->name);
427 if (err != 0) {
428 bp2->bio_error = EILSEQ;
429 DPRINTF(GUZ_DBG_ERR, ("%s/%s: rewind(%p) "
430 "failed\n", __func__, gp->name, sc->dcp));
431 goto done;
432 }
433 data += len;
434 } else
435 break;
436
437 data2 += ulen;
438 bp2->bio_completed += ulen;
439 bp2->bio_resid -= ulen;
440 bp->bio_completed -= len;
441 blkofs = 0;
442 blk++;
443 }
444
445 done:
446 /* Finish processing the request. */
447 free(bp->bio_data, M_GEOM_UZIP);
448 g_destroy_bio(bp);
449 if (bp2->bio_error != 0 || bp2->bio_resid == 0)
450 g_io_deliver(bp2, bp2->bio_error);
451 else
452 g_uzip_request(gp, bp2);
453 }
454
455 static void
g_uzip_start(struct bio * bp)456 g_uzip_start(struct bio *bp)
457 {
458 struct g_provider *pp;
459 struct g_geom *gp;
460 struct g_uzip_softc *sc;
461
462 pp = bp->bio_to;
463 gp = pp->geom;
464
465 DPRINTF(GUZ_DBG_IO, ("%s/%s: %p: cmd=%d, offset=%jd, length=%jd, "
466 "buffer=%p\n", __func__, gp->name, bp, bp->bio_cmd,
467 (intmax_t)bp->bio_offset, (intmax_t)bp->bio_length, bp->bio_data));
468
469 sc = gp->softc;
470 sc->req_total++;
471
472 if (bp->bio_cmd == BIO_GETATTR) {
473 struct bio *bp2;
474 struct g_consumer *cp;
475 struct g_geom *gp;
476 struct g_provider *pp;
477
478 /* pass on MNT:* requests and ignore others */
479 if (strncmp(bp->bio_attribute, "MNT:", 4) == 0) {
480 bp2 = g_clone_bio(bp);
481 if (bp2 == NULL) {
482 g_io_deliver(bp, ENOMEM);
483 return;
484 }
485 bp2->bio_done = g_std_done;
486 pp = bp->bio_to;
487 gp = pp->geom;
488 cp = LIST_FIRST(&gp->consumer);
489 g_io_request(bp2, cp);
490 return;
491 }
492 }
493 if (bp->bio_cmd != BIO_READ) {
494 g_io_deliver(bp, EOPNOTSUPP);
495 return;
496 }
497
498 bp->bio_resid = bp->bio_length;
499 bp->bio_completed = 0;
500
501 g_uzip_request(gp, bp);
502 }
503
504 static void
g_uzip_orphan(struct g_consumer * cp)505 g_uzip_orphan(struct g_consumer *cp)
506 {
507 struct g_geom *gp;
508
509 g_topology_assert();
510 G_VALID_CONSUMER(cp);
511 gp = cp->geom;
512 g_trace(G_T_TOPOLOGY, "%s(%p/%s)", __func__, cp, gp->name);
513 g_wither_geom(gp, ENXIO);
514
515 /*
516 * We can safely free the softc now if there are no accesses,
517 * otherwise g_uzip_access() will do that after the last close.
518 */
519 if ((cp->acr + cp->acw + cp->ace) == 0)
520 g_uzip_softc_free(gp);
521 }
522
523 static void
g_uzip_spoiled(struct g_consumer * cp)524 g_uzip_spoiled(struct g_consumer *cp)
525 {
526
527 g_trace(G_T_TOPOLOGY, "%s(%p/%s)", __func__, cp, cp->geom->name);
528 cp->flags |= G_CF_ORPHAN;
529 g_uzip_orphan(cp);
530 }
531
532 static int
g_uzip_access(struct g_provider * pp,int dr,int dw,int de)533 g_uzip_access(struct g_provider *pp, int dr, int dw, int de)
534 {
535 struct g_geom *gp;
536 struct g_consumer *cp;
537 int error;
538
539 gp = pp->geom;
540 cp = LIST_FIRST(&gp->consumer);
541 KASSERT (cp != NULL, ("g_uzip_access but no consumer"));
542
543 if (cp->acw + dw > 0)
544 return (EROFS);
545
546 error = g_access(cp, dr, dw, de);
547
548 /*
549 * Free the softc if all providers have been closed and this geom
550 * is being removed.
551 */
552 if (error == 0 && (gp->flags & G_GEOM_WITHER) != 0 &&
553 (cp->acr + cp->acw + cp->ace) == 0)
554 g_uzip_softc_free(gp);
555
556 return (error);
557 }
558
559 static int
g_uzip_parse_toc(struct g_uzip_softc * sc,struct g_provider * pp,struct g_geom * gp)560 g_uzip_parse_toc(struct g_uzip_softc *sc, struct g_provider *pp,
561 struct g_geom *gp)
562 {
563 uint32_t i, j, backref_to;
564 uint64_t max_offset, min_offset;
565 struct g_uzip_blk *last_blk;
566
567 min_offset = sizeof(struct cloop_header) +
568 (sc->nblocks + 1) * sizeof(uint64_t);
569 max_offset = sc->toc[0].offset - 1;
570 last_blk = &sc->toc[0];
571 for (i = 0; i < sc->nblocks; i++) {
572 /* First do some bounds checking */
573 if ((sc->toc[i].offset < min_offset) ||
574 (sc->toc[i].offset > pp->mediasize)) {
575 goto error_offset;
576 }
577 DPRINTF_BLK(GUZ_DBG_IO, i, ("%s: cluster #%u "
578 "offset=%ju max_offset=%ju\n", gp->name,
579 (u_int)i, (uintmax_t)sc->toc[i].offset,
580 (uintmax_t)max_offset));
581 backref_to = BLEN_UNDEF;
582 if (sc->toc[i].offset < max_offset) {
583 /*
584 * For the backref'ed blocks search already parsed
585 * TOC entries for the matching offset and copy the
586 * size from matched entry.
587 */
588 for (j = 0; j <= i; j++) {
589 if (sc->toc[j].offset == sc->toc[i].offset &&
590 !BLK_IS_NIL(sc, j)) {
591 break;
592 }
593 if (j != i) {
594 continue;
595 }
596 DPRINTF(GUZ_DBG_ERR, ("%s: cannot match "
597 "backref'ed offset at cluster #%u\n",
598 gp->name, i));
599 return (-1);
600 }
601 sc->toc[i].blen = sc->toc[j].blen;
602 backref_to = j;
603 } else {
604 last_blk = &sc->toc[i];
605 /*
606 * For the "normal blocks" seek forward until we hit
607 * block whose offset is larger than ours and assume
608 * it's going to be the next one.
609 */
610 for (j = i + 1; j < sc->nblocks + 1; j++) {
611 if (sc->toc[j].offset > max_offset) {
612 break;
613 }
614 }
615 sc->toc[i].blen = sc->toc[j].offset -
616 sc->toc[i].offset;
617 if (BLK_ENDS(sc, i) > pp->mediasize) {
618 DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u "
619 "extends past media boundary (%ju > %ju)\n",
620 gp->name, (u_int)i,
621 (uintmax_t)BLK_ENDS(sc, i),
622 (intmax_t)pp->mediasize));
623 return (-1);
624 }
625 KASSERT(max_offset <= sc->toc[i].offset, (
626 "%s: max_offset is incorrect: %ju",
627 gp->name, (uintmax_t)max_offset));
628 max_offset = BLK_ENDS(sc, i) - 1;
629 }
630 DPRINTF_BLK(GUZ_DBG_TOC, i, ("%s: cluster #%u, original %u "
631 "bytes, in %u bytes", gp->name, i, sc->blksz,
632 sc->toc[i].blen));
633 if (backref_to != BLEN_UNDEF) {
634 DPRINTF_BLK(GUZ_DBG_TOC, i, (" (->#%u)",
635 (u_int)backref_to));
636 }
637 DPRINTF_BLK(GUZ_DBG_TOC, i, ("\n"));
638 }
639 last_blk->last = 1;
640 /* Do a second pass to validate block lengths */
641 for (i = 0; i < sc->nblocks; i++) {
642 if (sc->toc[i].blen > sc->dcp->max_blen) {
643 if (sc->toc[i].last == 0) {
644 DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u "
645 "length (%ju) exceeds "
646 "max_blen (%ju)\n", gp->name, i,
647 (uintmax_t)sc->toc[i].blen,
648 (uintmax_t)sc->dcp->max_blen));
649 return (-1);
650 }
651 DPRINTF(GUZ_DBG_INFO, ("%s: cluster #%u extra "
652 "padding is detected, trimmed to %ju\n",
653 gp->name, i, (uintmax_t)sc->dcp->max_blen));
654 sc->toc[i].blen = sc->dcp->max_blen;
655 sc->toc[i].padded = 1;
656 }
657 }
658 return (0);
659
660 error_offset:
661 DPRINTF(GUZ_DBG_ERR, ("%s: cluster #%u: invalid offset %ju, "
662 "min_offset=%ju mediasize=%jd\n", gp->name, (u_int)i,
663 sc->toc[i].offset, min_offset, pp->mediasize));
664 return (-1);
665 }
666
667 static struct g_geom *
g_uzip_taste(struct g_class * mp,struct g_provider * pp,int flags)668 g_uzip_taste(struct g_class *mp, struct g_provider *pp, int flags)
669 {
670 int error;
671 uint32_t i, total_offsets, offsets_read, blk;
672 void *buf;
673 struct cloop_header *header;
674 struct g_consumer *cp;
675 struct g_geom *gp;
676 struct g_provider *pp2;
677 struct g_uzip_softc *sc;
678 struct g_geom_alias *gap;
679 enum {
680 G_UZIP = 1,
681 G_ULZMA,
682 G_ZSTD,
683 } type;
684 char cloop_version;
685
686 g_trace(G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name);
687 g_topology_assert();
688
689 /* Skip providers that are already open for writing. */
690 if (pp->acw > 0)
691 return (NULL);
692
693 if ((fnmatch(g_uzip_attach_to, pp->name, 0) != 0) ||
694 (fnmatch(g_uzip_noattach_to, pp->name, 0) == 0)) {
695 DPRINTF(GUZ_DBG_INFO, ("%s(%s,%s), ignoring\n", __func__,
696 mp->name, pp->name));
697 return (NULL);
698 }
699
700 buf = NULL;
701
702 /*
703 * Create geom instance.
704 */
705 gp = g_new_geomf(mp, GUZ_DEV_NAME("%s"), pp->name);
706 cp = g_new_consumer(gp);
707 error = g_attach(cp, pp);
708 if (error != 0)
709 goto e0;
710 error = g_access(cp, 1, 0, 0);
711 if (error)
712 goto e1;
713 g_topology_unlock();
714
715 /*
716 * Read cloop header, look for CLOOP magic, perform
717 * other validity checks.
718 */
719 DPRINTF(GUZ_DBG_INFO, ("%s: media sectorsize %u, mediasize %jd\n",
720 gp->name, pp->sectorsize, (intmax_t)pp->mediasize));
721 buf = g_read_data(cp, 0, pp->sectorsize, NULL);
722 if (buf == NULL)
723 goto e2;
724 header = (struct cloop_header *) buf;
725 if (strncmp(header->magic, CLOOP_MAGIC_START,
726 sizeof(CLOOP_MAGIC_START) - 1) != 0) {
727 DPRINTF(GUZ_DBG_ERR, ("%s: no CLOOP magic\n", gp->name));
728 goto e3;
729 }
730
731 cloop_version = header->magic[CLOOP_OFS_VERSN];
732 switch (header->magic[CLOOP_OFS_COMPR]) {
733 case CLOOP_COMP_LZMA:
734 case CLOOP_COMP_LZMA_DDP:
735 type = G_ULZMA;
736 if (cloop_version < CLOOP_MINVER_LZMA) {
737 DPRINTF(GUZ_DBG_ERR, ("%s: image version too old\n",
738 gp->name));
739 goto e3;
740 }
741 DPRINTF(GUZ_DBG_INFO, ("%s: GEOM_UZIP_LZMA image found\n",
742 gp->name));
743 break;
744 case CLOOP_COMP_LIBZ:
745 case CLOOP_COMP_LIBZ_DDP:
746 type = G_UZIP;
747 if (cloop_version < CLOOP_MINVER_ZLIB) {
748 DPRINTF(GUZ_DBG_ERR, ("%s: image version too old\n",
749 gp->name));
750 goto e3;
751 }
752 DPRINTF(GUZ_DBG_INFO, ("%s: GEOM_UZIP_ZLIB image found\n",
753 gp->name));
754 break;
755 case CLOOP_COMP_ZSTD:
756 case CLOOP_COMP_ZSTD_DDP:
757 if (cloop_version < CLOOP_MINVER_ZSTD) {
758 DPRINTF(GUZ_DBG_ERR, ("%s: image version too old\n",
759 gp->name));
760 goto e3;
761 }
762 #ifdef ZSTDIO
763 DPRINTF(GUZ_DBG_INFO, ("%s: GEOM_UZIP_ZSTD image found.\n",
764 gp->name));
765 type = G_ZSTD;
766 #else
767 DPRINTF(GUZ_DBG_ERR, ("%s: GEOM_UZIP_ZSTD image found, but "
768 "this kernel was configured with Zstd disabled.\n",
769 gp->name));
770 goto e3;
771 #endif
772 break;
773 default:
774 DPRINTF(GUZ_DBG_ERR, ("%s: unsupported image type\n",
775 gp->name));
776 goto e3;
777 }
778
779 /*
780 * Initialize softc and read offsets.
781 */
782 sc = malloc(sizeof(*sc), M_GEOM_UZIP, M_WAITOK | M_ZERO);
783 gp->softc = sc;
784 sc->blksz = ntohl(header->blksz);
785 sc->nblocks = ntohl(header->nblocks);
786 if (sc->blksz % 512 != 0) {
787 printf("%s: block size (%u) should be multiple of 512.\n",
788 gp->name, sc->blksz);
789 goto e4;
790 }
791 if (sc->blksz > MAX_BLKSZ) {
792 printf("%s: block size (%u) should not be larger than %lu.\n",
793 gp->name, sc->blksz, MAX_BLKSZ);
794 }
795 total_offsets = sc->nblocks + 1;
796 if (sizeof(struct cloop_header) +
797 total_offsets * sizeof(uint64_t) > pp->mediasize) {
798 printf("%s: media too small for %u blocks\n",
799 gp->name, sc->nblocks);
800 goto e4;
801 }
802 sc->toc = malloc(total_offsets * sizeof(struct g_uzip_blk),
803 M_GEOM_UZIP, M_WAITOK | M_ZERO);
804 offsets_read = MIN(total_offsets,
805 (pp->sectorsize - sizeof(*header)) / sizeof(uint64_t));
806 for (i = 0; i < offsets_read; i++) {
807 sc->toc[i].offset = be64toh(((uint64_t *) (header + 1))[i]);
808 sc->toc[i].blen = BLEN_UNDEF;
809 }
810 DPRINTF(GUZ_DBG_INFO, ("%s: %u offsets in the first sector\n",
811 gp->name, offsets_read));
812
813 /*
814 * The following invalidates the "header" pointer into the first
815 * block's "buf."
816 */
817 header = NULL;
818
819 for (blk = 1; offsets_read < total_offsets; blk++) {
820 uint32_t nread;
821
822 free(buf, M_GEOM);
823 buf = g_read_data(
824 cp, blk * pp->sectorsize, pp->sectorsize, NULL);
825 if (buf == NULL)
826 goto e5;
827 nread = MIN(total_offsets - offsets_read,
828 pp->sectorsize / sizeof(uint64_t));
829 DPRINTF(GUZ_DBG_TOC, ("%s: %u offsets read from sector %d\n",
830 gp->name, nread, blk));
831 for (i = 0; i < nread; i++) {
832 sc->toc[offsets_read + i].offset =
833 be64toh(((uint64_t *) buf)[i]);
834 sc->toc[offsets_read + i].blen = BLEN_UNDEF;
835 }
836 offsets_read += nread;
837 }
838 free(buf, M_GEOM);
839 buf = NULL;
840 offsets_read -= 1;
841 DPRINTF(GUZ_DBG_INFO, ("%s: done reading %u block offsets from %u "
842 "sectors\n", gp->name, offsets_read, blk));
843 if (sc->nblocks != offsets_read) {
844 DPRINTF(GUZ_DBG_ERR, ("%s: read %s offsets than expected "
845 "blocks\n", gp->name,
846 sc->nblocks < offsets_read ? "more" : "less"));
847 goto e5;
848 }
849
850 switch (type) {
851 case G_UZIP:
852 sc->dcp = g_uzip_zlib_ctor(sc->blksz);
853 break;
854 case G_ULZMA:
855 sc->dcp = g_uzip_lzma_ctor(sc->blksz);
856 break;
857 #ifdef ZSTDIO
858 case G_ZSTD:
859 sc->dcp = g_uzip_zstd_ctor(sc->blksz);
860 break;
861 #endif
862 default:
863 goto e5;
864 }
865
866 /*
867 * The last+1 block was not always initialized by earlier versions of
868 * mkuzip(8). However, *if* it is initialized, the difference between
869 * its offset and the prior block's offset represents the length of the
870 * final real compressed block, and this is significant to the
871 * decompressor.
872 */
873 if (cloop_version >= CLOOP_MINVER_RELIABLE_LASTBLKSZ &&
874 sc->toc[sc->nblocks].offset != 0) {
875 if (sc->toc[sc->nblocks].offset > pp->mediasize) {
876 DPRINTF(GUZ_DBG_ERR,
877 ("%s: bogus n+1 offset %ju > mediasize %ju\n",
878 gp->name, (uintmax_t)sc->toc[sc->nblocks].offset,
879 (uintmax_t)pp->mediasize));
880 goto e6;
881 }
882 } else {
883 sc->toc[sc->nblocks].offset = pp->mediasize;
884 }
885 /* Massage TOC (table of contents), make sure it is sound */
886 if (g_uzip_parse_toc(sc, pp, gp) != 0) {
887 DPRINTF(GUZ_DBG_ERR, ("%s: TOC error\n", gp->name));
888 goto e6;
889 }
890 mtx_init(&sc->last_mtx, "geom_uzip cache", NULL, MTX_DEF);
891 mtx_init(&sc->queue_mtx, "geom_uzip wrkthread", NULL, MTX_DEF);
892 bioq_init(&sc->bio_queue);
893 sc->last_blk = -1;
894 sc->last_buf = malloc(sc->blksz, M_GEOM_UZIP, M_WAITOK);
895 sc->req_total = 0;
896 sc->req_cached = 0;
897
898 sc->uzip_do = &g_uzip_do;
899
900 error = kproc_create(g_uzip_wrkthr, sc, &sc->procp, 0, 0, "%s",
901 gp->name);
902 if (error != 0) {
903 goto e7;
904 }
905
906 g_topology_lock();
907 pp2 = g_new_providerf(gp, "%s", gp->name);
908 pp2->sectorsize = 512;
909 pp2->mediasize = (off_t)sc->nblocks * sc->blksz;
910 pp2->stripesize = pp->stripesize;
911 pp2->stripeoffset = pp->stripeoffset;
912 LIST_FOREACH(gap, &pp->aliases, ga_next)
913 g_provider_add_alias(pp2, GUZ_DEV_NAME("%s"), gap->ga_alias);
914 g_error_provider(pp2, 0);
915 g_access(cp, -1, 0, 0);
916
917 DPRINTF(GUZ_DBG_INFO, ("%s: taste ok (%d, %ju), (%ju, %ju), %x\n",
918 gp->name, pp2->sectorsize, (uintmax_t)pp2->mediasize,
919 (uintmax_t)pp2->stripeoffset, (uintmax_t)pp2->stripesize, pp2->flags));
920 DPRINTF(GUZ_DBG_INFO, ("%s: %u x %u blocks\n", gp->name, sc->nblocks,
921 sc->blksz));
922 return (gp);
923
924 e7:
925 free(sc->last_buf, M_GEOM);
926 mtx_destroy(&sc->queue_mtx);
927 mtx_destroy(&sc->last_mtx);
928 e6:
929 sc->dcp->free(sc->dcp);
930 e5:
931 free(sc->toc, M_GEOM);
932 e4:
933 free(gp->softc, M_GEOM_UZIP);
934 e3:
935 if (buf != NULL) {
936 free(buf, M_GEOM);
937 }
938 e2:
939 g_topology_lock();
940 g_access(cp, -1, 0, 0);
941 e1:
942 g_detach(cp);
943 e0:
944 g_destroy_consumer(cp);
945 g_destroy_geom(gp);
946
947 return (NULL);
948 }
949
950 static int
g_uzip_destroy_geom(struct gctl_req * req,struct g_class * mp,struct g_geom * gp)951 g_uzip_destroy_geom(struct gctl_req *req, struct g_class *mp, struct g_geom *gp)
952 {
953 struct g_provider *pp;
954
955 KASSERT(gp != NULL, ("NULL geom"));
956 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, gp->name);
957 g_topology_assert();
958
959 if (gp->softc == NULL) {
960 DPRINTF(GUZ_DBG_ERR, ("%s(%s): gp->softc == NULL\n", __func__,
961 gp->name));
962 return (ENXIO);
963 }
964
965 pp = LIST_FIRST(&gp->provider);
966 KASSERT(pp != NULL, ("NULL provider"));
967 if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)
968 return (EBUSY);
969
970 g_wither_geom(gp, ENXIO);
971 g_uzip_softc_free(gp);
972 return (0);
973 }
974
975 static struct g_class g_uzip_class = {
976 .name = UZIP_CLASS_NAME,
977 .version = G_VERSION,
978 .taste = g_uzip_taste,
979 .destroy_geom = g_uzip_destroy_geom,
980
981 .start = g_uzip_start,
982 .orphan = g_uzip_orphan,
983 .access = g_uzip_access,
984 .spoiled = g_uzip_spoiled,
985 };
986
987 DECLARE_GEOM_CLASS(g_uzip_class, g_uzip);
988 MODULE_DEPEND(g_uzip, xz, 1, 1, 1);
989 MODULE_DEPEND(g_uzip, zlib, 1, 1, 1);
990 MODULE_VERSION(geom_uzip, 0);
991