xref: /freebsd/sys/kern/subr_compressor.c (revision 685dc743dc3b5645e34836464128e1c0558b404b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2014, 2017 Mark Johnston <markj@FreeBSD.org>
5  * Copyright (c) 2017 Conrad Meyer <cem@FreeBSD.org>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions are
9  * 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
14  *    the documentation and/or other materials provided with the
15  *    distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * Subroutines used for writing compressed user process and kernel core dumps.
32  */
33 
34 #include <sys/cdefs.h>
35 #include "opt_gzio.h"
36 #include "opt_zstdio.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 
41 #include <sys/compressor.h>
42 #include <sys/endian.h>
43 #include <sys/kernel.h>
44 #include <sys/linker_set.h>
45 #include <sys/malloc.h>
46 
47 MALLOC_DEFINE(M_COMPRESS, "compressor", "kernel compression subroutines");
48 
49 struct compressor_methods {
50 	int format;
51 	void *(* const init)(size_t, int);
52 	void (* const reset)(void *);
53 	int (* const write)(void *, void *, size_t, compressor_cb_t, void *);
54 	void (* const fini)(void *);
55 };
56 
57 struct compressor {
58 	const struct compressor_methods *methods;
59 	compressor_cb_t cb;
60 	void *priv;
61 	void *arg;
62 };
63 
64 SET_DECLARE(compressors, struct compressor_methods);
65 
66 #ifdef GZIO
67 
68 #include <contrib/zlib/zutil.h>
69 
70 struct gz_stream {
71 	uint8_t		*gz_buffer;	/* output buffer */
72 	size_t		gz_bufsz;	/* output buffer size */
73 	off_t		gz_off;		/* offset into the output stream */
74 	uint32_t	gz_crc;		/* stream CRC32 */
75 	z_stream	gz_stream;	/* zlib state */
76 };
77 
78 static void 	*gz_init(size_t maxiosize, int level);
79 static void	gz_reset(void *stream);
80 static int	gz_write(void *stream, void *data, size_t len, compressor_cb_t,
81 		    void *);
82 static void	gz_fini(void *stream);
83 
84 static void *
gz_alloc(void * arg __unused,u_int n,u_int sz)85 gz_alloc(void *arg __unused, u_int n, u_int sz)
86 {
87 
88 	/*
89 	 * Memory for zlib state is allocated using M_NODUMP since it may be
90 	 * used to compress a kernel dump, and we don't want zlib to attempt to
91 	 * compress its own state.
92 	 */
93 	return (malloc(n * sz, M_COMPRESS, M_WAITOK | M_ZERO | M_NODUMP));
94 }
95 
96 static void
gz_free(void * arg __unused,void * ptr)97 gz_free(void *arg __unused, void *ptr)
98 {
99 
100 	free(ptr, M_COMPRESS);
101 }
102 
103 static void *
gz_init(size_t maxiosize,int level)104 gz_init(size_t maxiosize, int level)
105 {
106 	struct gz_stream *s;
107 	int error;
108 
109 	s = gz_alloc(NULL, 1, roundup2(sizeof(*s), PAGE_SIZE));
110 	s->gz_buffer = gz_alloc(NULL, 1, maxiosize);
111 	s->gz_bufsz = maxiosize;
112 
113 	s->gz_stream.zalloc = gz_alloc;
114 	s->gz_stream.zfree = gz_free;
115 	s->gz_stream.opaque = NULL;
116 	s->gz_stream.next_in = Z_NULL;
117 	s->gz_stream.avail_in = 0;
118 
119 	if (level != Z_DEFAULT_COMPRESSION) {
120 		if (level < Z_BEST_SPEED)
121 			level = Z_BEST_SPEED;
122 		else if (level > Z_BEST_COMPRESSION)
123 			level = Z_BEST_COMPRESSION;
124 	}
125 
126 	error = deflateInit2(&s->gz_stream, level, Z_DEFLATED, -MAX_WBITS,
127 	    DEF_MEM_LEVEL, Z_DEFAULT_STRATEGY);
128 	if (error != 0)
129 		goto fail;
130 
131 	gz_reset(s);
132 
133 	return (s);
134 
135 fail:
136 	gz_free(NULL, s);
137 	return (NULL);
138 }
139 
140 static void
gz_reset(void * stream)141 gz_reset(void *stream)
142 {
143 	struct gz_stream *s;
144 	uint8_t *hdr;
145 	const size_t hdrlen = 10;
146 
147 	s = stream;
148 	s->gz_off = 0;
149 	s->gz_crc = crc32(0L, Z_NULL, 0);
150 
151 	(void)deflateReset(&s->gz_stream);
152 	s->gz_stream.avail_out = s->gz_bufsz;
153 	s->gz_stream.next_out = s->gz_buffer;
154 
155 	/* Write the gzip header to the output buffer. */
156 	hdr = s->gz_buffer;
157 	memset(hdr, 0, hdrlen);
158 	hdr[0] = 0x1f;
159 	hdr[1] = 0x8b;
160 	hdr[2] = Z_DEFLATED;
161 	hdr[9] = OS_CODE;
162 	s->gz_stream.next_out += hdrlen;
163 	s->gz_stream.avail_out -= hdrlen;
164 }
165 
166 static int
gz_write(void * stream,void * data,size_t len,compressor_cb_t cb,void * arg)167 gz_write(void *stream, void *data, size_t len, compressor_cb_t cb,
168     void *arg)
169 {
170 	struct gz_stream *s;
171 	uint8_t trailer[8];
172 	size_t room;
173 	int error, zerror, zflag;
174 
175 	s = stream;
176 	zflag = data == NULL ? Z_FINISH : Z_NO_FLUSH;
177 
178 	if (len > 0) {
179 		s->gz_stream.avail_in = len;
180 		s->gz_stream.next_in = data;
181 		s->gz_crc = crc32(s->gz_crc, data, len);
182 	}
183 
184 	error = 0;
185 	do {
186 		zerror = deflate(&s->gz_stream, zflag);
187 		if (zerror != Z_OK && zerror != Z_STREAM_END) {
188 			error = EIO;
189 			break;
190 		}
191 
192 		if (s->gz_stream.avail_out == 0 || zerror == Z_STREAM_END) {
193 			/*
194 			 * Our output buffer is full or there's nothing left
195 			 * to produce, so we're flushing the buffer.
196 			 */
197 			len = s->gz_bufsz - s->gz_stream.avail_out;
198 			if (zerror == Z_STREAM_END) {
199 				/*
200 				 * Try to pack as much of the trailer into the
201 				 * output buffer as we can.
202 				 */
203 				((uint32_t *)trailer)[0] = htole32(s->gz_crc);
204 				((uint32_t *)trailer)[1] =
205 				    htole32(s->gz_stream.total_in);
206 				room = MIN(sizeof(trailer),
207 				    s->gz_bufsz - len);
208 				memcpy(s->gz_buffer + len, trailer, room);
209 				len += room;
210 			}
211 
212 			error = cb(s->gz_buffer, len, s->gz_off, arg);
213 			if (error != 0)
214 				break;
215 
216 			s->gz_off += len;
217 			s->gz_stream.next_out = s->gz_buffer;
218 			s->gz_stream.avail_out = s->gz_bufsz;
219 
220 			/*
221 			 * If we couldn't pack the trailer into the output
222 			 * buffer, write it out now.
223 			 */
224 			if (zerror == Z_STREAM_END && room < sizeof(trailer))
225 				error = cb(trailer + room,
226 				    sizeof(trailer) - room, s->gz_off, arg);
227 		}
228 	} while (zerror != Z_STREAM_END &&
229 	    (zflag == Z_FINISH || s->gz_stream.avail_in > 0));
230 
231 	return (error);
232 }
233 
234 static void
gz_fini(void * stream)235 gz_fini(void *stream)
236 {
237 	struct gz_stream *s;
238 
239 	s = stream;
240 	(void)deflateEnd(&s->gz_stream);
241 	gz_free(NULL, s->gz_buffer);
242 	gz_free(NULL, s);
243 }
244 
245 struct compressor_methods gzip_methods = {
246 	.format = COMPRESS_GZIP,
247 	.init = gz_init,
248 	.reset = gz_reset,
249 	.write = gz_write,
250 	.fini = gz_fini,
251 };
252 DATA_SET(compressors, gzip_methods);
253 
254 #endif /* GZIO */
255 
256 #ifdef ZSTDIO
257 
258 #define	ZSTD_STATIC_LINKING_ONLY
259 #include <contrib/zstd/lib/zstd.h>
260 
261 struct zstdio_stream {
262 	ZSTD_CCtx	*zst_stream;
263 	ZSTD_inBuffer	zst_inbuffer;
264 	ZSTD_outBuffer	zst_outbuffer;
265 	uint8_t *	zst_buffer;	/* output buffer */
266 	size_t		zst_maxiosz;	/* Max output IO size */
267 	off_t		zst_off;	/* offset into the output stream */
268 	void *		zst_static_wkspc;
269 };
270 
271 static void 	*zstdio_init(size_t maxiosize, int level);
272 static void	zstdio_reset(void *stream);
273 static int	zstdio_write(void *stream, void *data, size_t len,
274 		    compressor_cb_t, void *);
275 static void	zstdio_fini(void *stream);
276 
277 static void *
zstdio_init(size_t maxiosize,int level)278 zstdio_init(size_t maxiosize, int level)
279 {
280 	ZSTD_CCtx *dump_compressor;
281 	struct zstdio_stream *s;
282 	void *wkspc, *owkspc, *buffer;
283 	size_t wkspc_size, buf_size, rc;
284 
285 	s = NULL;
286 	wkspc_size = ZSTD_estimateCStreamSize(level);
287 	owkspc = wkspc = malloc(wkspc_size + 8, M_COMPRESS,
288 	    M_WAITOK | M_NODUMP);
289 	/* Zstd API requires 8-byte alignment. */
290 	if ((uintptr_t)wkspc % 8 != 0)
291 		wkspc = (void *)roundup2((uintptr_t)wkspc, 8);
292 
293 	dump_compressor = ZSTD_initStaticCCtx(wkspc, wkspc_size);
294 	if (dump_compressor == NULL) {
295 		printf("%s: workspace too small.\n", __func__);
296 		goto out;
297 	}
298 
299 	rc = ZSTD_CCtx_setParameter(dump_compressor, ZSTD_c_checksumFlag, 1);
300 	if (ZSTD_isError(rc)) {
301 		printf("%s: error setting checksumFlag: %s\n", __func__,
302 		    ZSTD_getErrorName(rc));
303 		goto out;
304 	}
305 	rc = ZSTD_CCtx_setParameter(dump_compressor, ZSTD_c_compressionLevel,
306 	    level);
307 	if (ZSTD_isError(rc)) {
308 		printf("%s: error setting compressLevel: %s\n", __func__,
309 		    ZSTD_getErrorName(rc));
310 		goto out;
311 	}
312 
313 	buf_size = ZSTD_CStreamOutSize() * 2;
314 	buffer = malloc(buf_size, M_COMPRESS, M_WAITOK | M_NODUMP);
315 
316 	s = malloc(sizeof(*s), M_COMPRESS, M_NODUMP | M_WAITOK);
317 	s->zst_buffer = buffer;
318 	s->zst_outbuffer.dst = buffer;
319 	s->zst_outbuffer.size = buf_size;
320 	s->zst_maxiosz = maxiosize;
321 	s->zst_stream = dump_compressor;
322 	s->zst_static_wkspc = owkspc;
323 
324 	zstdio_reset(s);
325 
326 out:
327 	if (s == NULL)
328 		free(owkspc, M_COMPRESS);
329 	return (s);
330 }
331 
332 static void
zstdio_reset(void * stream)333 zstdio_reset(void *stream)
334 {
335 	struct zstdio_stream *s;
336 	size_t res;
337 
338 	s = stream;
339 	res = ZSTD_CCtx_reset(s->zst_stream, ZSTD_reset_session_only);
340 	if (ZSTD_isError(res))
341 		panic("%s: could not reset stream %p: %s\n", __func__, s,
342 		    ZSTD_getErrorName(res));
343 	res = ZSTD_CCtx_setPledgedSrcSize(s->zst_stream,
344 	    ZSTD_CONTENTSIZE_UNKNOWN);
345 	if (ZSTD_isError(res))
346 		panic("%s: could not set src size on %p: %s\n", __func__, s,
347 		    ZSTD_getErrorName(res));
348 
349 	s->zst_off = 0;
350 	s->zst_inbuffer.src = NULL;
351 	s->zst_inbuffer.size = 0;
352 	s->zst_inbuffer.pos = 0;
353 	s->zst_outbuffer.pos = 0;
354 }
355 
356 static int
zst_flush_intermediate(struct zstdio_stream * s,compressor_cb_t cb,void * arg)357 zst_flush_intermediate(struct zstdio_stream *s, compressor_cb_t cb, void *arg)
358 {
359 	size_t bytes_to_dump;
360 	int error;
361 
362 	/* Flush as many full output blocks as possible. */
363 	/* XXX: 4096 is arbitrary safe HDD block size for kernel dumps */
364 	while (s->zst_outbuffer.pos >= 4096) {
365 		bytes_to_dump = rounddown(s->zst_outbuffer.pos, 4096);
366 
367 		if (bytes_to_dump > s->zst_maxiosz)
368 			bytes_to_dump = s->zst_maxiosz;
369 
370 		error = cb(s->zst_buffer, bytes_to_dump, s->zst_off, arg);
371 		if (error != 0)
372 			return (error);
373 
374 		/*
375 		 * Shift any non-full blocks up to the front of the output
376 		 * buffer.
377 		 */
378 		s->zst_outbuffer.pos -= bytes_to_dump;
379 		memmove(s->zst_outbuffer.dst,
380 		    (char *)s->zst_outbuffer.dst + bytes_to_dump,
381 		    s->zst_outbuffer.pos);
382 		s->zst_off += bytes_to_dump;
383 	}
384 	return (0);
385 }
386 
387 static int
zstdio_flush(struct zstdio_stream * s,compressor_cb_t cb,void * arg)388 zstdio_flush(struct zstdio_stream *s, compressor_cb_t cb, void *arg)
389 {
390 	size_t rc, lastpos;
391 	int error;
392 
393 	/*
394 	 * Positive return indicates unflushed data remaining; need to call
395 	 * endStream again after clearing out room in output buffer.
396 	 */
397 	rc = 1;
398 	lastpos = s->zst_outbuffer.pos;
399 	while (rc > 0) {
400 		rc = ZSTD_endStream(s->zst_stream, &s->zst_outbuffer);
401 		if (ZSTD_isError(rc)) {
402 			printf("%s: ZSTD_endStream failed (%s)\n", __func__,
403 			    ZSTD_getErrorName(rc));
404 			return (EIO);
405 		}
406 		if (lastpos == s->zst_outbuffer.pos) {
407 			printf("%s: did not make forward progress endStream %zu\n",
408 			    __func__, lastpos);
409 			return (EIO);
410 		}
411 
412 		error = zst_flush_intermediate(s, cb, arg);
413 		if (error != 0)
414 			return (error);
415 
416 		lastpos = s->zst_outbuffer.pos;
417 	}
418 
419 	/*
420 	 * We've already done an intermediate flush, so all full blocks have
421 	 * been written.  Only a partial block remains.  Padding happens in a
422 	 * higher layer.
423 	 */
424 	if (s->zst_outbuffer.pos != 0) {
425 		error = cb(s->zst_buffer, s->zst_outbuffer.pos, s->zst_off,
426 		    arg);
427 		if (error != 0)
428 			return (error);
429 	}
430 
431 	return (0);
432 }
433 
434 static int
zstdio_write(void * stream,void * data,size_t len,compressor_cb_t cb,void * arg)435 zstdio_write(void *stream, void *data, size_t len, compressor_cb_t cb,
436     void *arg)
437 {
438 	struct zstdio_stream *s;
439 	size_t lastpos, rc;
440 	int error;
441 
442 	s = stream;
443 	if (data == NULL)
444 		return (zstdio_flush(s, cb, arg));
445 
446 	s->zst_inbuffer.src = data;
447 	s->zst_inbuffer.size = len;
448 	s->zst_inbuffer.pos = 0;
449 	lastpos = 0;
450 
451 	while (s->zst_inbuffer.pos < s->zst_inbuffer.size) {
452 		rc = ZSTD_compressStream(s->zst_stream, &s->zst_outbuffer,
453 		    &s->zst_inbuffer);
454 		if (ZSTD_isError(rc)) {
455 			printf("%s: Compress failed on %p! (%s)\n",
456 			    __func__, data, ZSTD_getErrorName(rc));
457 			return (EIO);
458 		}
459 
460 		if (lastpos == s->zst_inbuffer.pos) {
461 			/*
462 			 * XXX: May need flushStream to make forward progress
463 			 */
464 			printf("ZSTD: did not make forward progress @pos %zu\n",
465 			    lastpos);
466 			return (EIO);
467 		}
468 		lastpos = s->zst_inbuffer.pos;
469 
470 		error = zst_flush_intermediate(s, cb, arg);
471 		if (error != 0)
472 			return (error);
473 	}
474 	return (0);
475 }
476 
477 static void
zstdio_fini(void * stream)478 zstdio_fini(void *stream)
479 {
480 	struct zstdio_stream *s;
481 
482 	s = stream;
483 	if (s->zst_static_wkspc != NULL)
484 		free(s->zst_static_wkspc, M_COMPRESS);
485 	else
486 		ZSTD_freeCCtx(s->zst_stream);
487 	free(s->zst_buffer, M_COMPRESS);
488 	free(s, M_COMPRESS);
489 }
490 
491 static struct compressor_methods zstd_methods = {
492 	.format = COMPRESS_ZSTD,
493 	.init = zstdio_init,
494 	.reset = zstdio_reset,
495 	.write = zstdio_write,
496 	.fini = zstdio_fini,
497 };
498 DATA_SET(compressors, zstd_methods);
499 
500 #endif /* ZSTDIO */
501 
502 bool
compressor_avail(int format)503 compressor_avail(int format)
504 {
505 	struct compressor_methods **iter;
506 
507 	SET_FOREACH(iter, compressors) {
508 		if ((*iter)->format == format)
509 			return (true);
510 	}
511 	return (false);
512 }
513 
514 struct compressor *
compressor_init(compressor_cb_t cb,int format,size_t maxiosize,int level,void * arg)515 compressor_init(compressor_cb_t cb, int format, size_t maxiosize, int level,
516     void *arg)
517 {
518 	struct compressor_methods **iter;
519 	struct compressor *s;
520 	void *priv;
521 
522 	SET_FOREACH(iter, compressors) {
523 		if ((*iter)->format == format)
524 			break;
525 	}
526 	if (iter == SET_LIMIT(compressors))
527 		return (NULL);
528 
529 	priv = (*iter)->init(maxiosize, level);
530 	if (priv == NULL)
531 		return (NULL);
532 
533 	s = malloc(sizeof(*s), M_COMPRESS, M_WAITOK | M_ZERO);
534 	s->methods = (*iter);
535 	s->priv = priv;
536 	s->cb = cb;
537 	s->arg = arg;
538 	return (s);
539 }
540 
541 void
compressor_reset(struct compressor * stream)542 compressor_reset(struct compressor *stream)
543 {
544 
545 	stream->methods->reset(stream->priv);
546 }
547 
548 int
compressor_write(struct compressor * stream,void * data,size_t len)549 compressor_write(struct compressor *stream, void *data, size_t len)
550 {
551 
552 	return (stream->methods->write(stream->priv, data, len, stream->cb,
553 	    stream->arg));
554 }
555 
556 int
compressor_flush(struct compressor * stream)557 compressor_flush(struct compressor *stream)
558 {
559 
560 	return (stream->methods->write(stream->priv, NULL, 0, stream->cb,
561 	    stream->arg));
562 }
563 
564 void
compressor_fini(struct compressor * stream)565 compressor_fini(struct compressor *stream)
566 {
567 
568 	stream->methods->fini(stream->priv);
569 }
570