1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2019 HUAWEI, Inc.
4 *             https://www.huawei.com/
5 */
6#include "compress.h"
7#include <linux/lz4.h>
8
9#ifndef LZ4_DISTANCE_MAX	/* history window size */
10#define LZ4_DISTANCE_MAX 65535	/* set to maximum value by default */
11#endif
12
13#define LZ4_MAX_DISTANCE_PAGES	(DIV_ROUND_UP(LZ4_DISTANCE_MAX, PAGE_SIZE) + 1)
14#ifndef LZ4_DECOMPRESS_INPLACE_MARGIN
15#define LZ4_DECOMPRESS_INPLACE_MARGIN(srcsize)  (((srcsize) >> 8) + 32)
16#endif
17
18struct z_erofs_lz4_decompress_ctx {
19	struct z_erofs_decompress_req *rq;
20	/* # of encoded, decoded pages */
21	unsigned int inpages, outpages;
22	/* decoded block total length (used for in-place decompression) */
23	unsigned int oend;
24};
25
26static int z_erofs_load_lz4_config(struct super_block *sb,
27			    struct erofs_super_block *dsb, void *data, int size)
28{
29	struct erofs_sb_info *sbi = EROFS_SB(sb);
30	struct z_erofs_lz4_cfgs *lz4 = data;
31	u16 distance;
32
33	if (lz4) {
34		if (size < sizeof(struct z_erofs_lz4_cfgs)) {
35			erofs_err(sb, "invalid lz4 cfgs, size=%u", size);
36			return -EINVAL;
37		}
38		distance = le16_to_cpu(lz4->max_distance);
39
40		sbi->lz4.max_pclusterblks = le16_to_cpu(lz4->max_pclusterblks);
41		if (!sbi->lz4.max_pclusterblks) {
42			sbi->lz4.max_pclusterblks = 1;	/* reserved case */
43		} else if (sbi->lz4.max_pclusterblks >
44			   erofs_blknr(sb, Z_EROFS_PCLUSTER_MAX_SIZE)) {
45			erofs_err(sb, "too large lz4 pclusterblks %u",
46				  sbi->lz4.max_pclusterblks);
47			return -EINVAL;
48		}
49	} else {
50		distance = le16_to_cpu(dsb->u1.lz4_max_distance);
51		sbi->lz4.max_pclusterblks = 1;
52	}
53
54	sbi->lz4.max_distance_pages = distance ?
55					DIV_ROUND_UP(distance, PAGE_SIZE) + 1 :
56					LZ4_MAX_DISTANCE_PAGES;
57	return erofs_pcpubuf_growsize(sbi->lz4.max_pclusterblks);
58}
59
60/*
61 * Fill all gaps with bounce pages if it's a sparse page list. Also check if
62 * all physical pages are consecutive, which can be seen for moderate CR.
63 */
64static int z_erofs_lz4_prepare_dstpages(struct z_erofs_lz4_decompress_ctx *ctx,
65					struct page **pagepool)
66{
67	struct z_erofs_decompress_req *rq = ctx->rq;
68	struct page *availables[LZ4_MAX_DISTANCE_PAGES] = { NULL };
69	unsigned long bounced[DIV_ROUND_UP(LZ4_MAX_DISTANCE_PAGES,
70					   BITS_PER_LONG)] = { 0 };
71	unsigned int lz4_max_distance_pages =
72				EROFS_SB(rq->sb)->lz4.max_distance_pages;
73	void *kaddr = NULL;
74	unsigned int i, j, top;
75
76	top = 0;
77	for (i = j = 0; i < ctx->outpages; ++i, ++j) {
78		struct page *const page = rq->out[i];
79		struct page *victim;
80
81		if (j >= lz4_max_distance_pages)
82			j = 0;
83
84		/* 'valid' bounced can only be tested after a complete round */
85		if (!rq->fillgaps && test_bit(j, bounced)) {
86			DBG_BUGON(i < lz4_max_distance_pages);
87			DBG_BUGON(top >= lz4_max_distance_pages);
88			availables[top++] = rq->out[i - lz4_max_distance_pages];
89		}
90
91		if (page) {
92			__clear_bit(j, bounced);
93			if (!PageHighMem(page)) {
94				if (!i) {
95					kaddr = page_address(page);
96					continue;
97				}
98				if (kaddr &&
99				    kaddr + PAGE_SIZE == page_address(page)) {
100					kaddr += PAGE_SIZE;
101					continue;
102				}
103			}
104			kaddr = NULL;
105			continue;
106		}
107		kaddr = NULL;
108		__set_bit(j, bounced);
109
110		if (top) {
111			victim = availables[--top];
112			get_page(victim);
113		} else {
114			victim = erofs_allocpage(pagepool, rq->gfp);
115			if (!victim)
116				return -ENOMEM;
117			set_page_private(victim, Z_EROFS_SHORTLIVED_PAGE);
118		}
119		rq->out[i] = victim;
120	}
121	return kaddr ? 1 : 0;
122}
123
124static void *z_erofs_lz4_handle_overlap(struct z_erofs_lz4_decompress_ctx *ctx,
125			void *inpage, void *out, unsigned int *inputmargin,
126			int *maptype, bool may_inplace)
127{
128	struct z_erofs_decompress_req *rq = ctx->rq;
129	unsigned int omargin, total, i;
130	struct page **in;
131	void *src, *tmp;
132
133	if (rq->inplace_io) {
134		omargin = PAGE_ALIGN(ctx->oend) - ctx->oend;
135		if (rq->partial_decoding || !may_inplace ||
136		    omargin < LZ4_DECOMPRESS_INPLACE_MARGIN(rq->inputsize))
137			goto docopy;
138
139		for (i = 0; i < ctx->inpages; ++i)
140			if (rq->out[ctx->outpages - ctx->inpages + i] !=
141			    rq->in[i])
142				goto docopy;
143		kunmap_local(inpage);
144		*maptype = 3;
145		return out + ((ctx->outpages - ctx->inpages) << PAGE_SHIFT);
146	}
147
148	if (ctx->inpages <= 1) {
149		*maptype = 0;
150		return inpage;
151	}
152	kunmap_local(inpage);
153	src = erofs_vm_map_ram(rq->in, ctx->inpages);
154	if (!src)
155		return ERR_PTR(-ENOMEM);
156	*maptype = 1;
157	return src;
158
159docopy:
160	/* Or copy compressed data which can be overlapped to per-CPU buffer */
161	in = rq->in;
162	src = erofs_get_pcpubuf(ctx->inpages);
163	if (!src) {
164		DBG_BUGON(1);
165		kunmap_local(inpage);
166		return ERR_PTR(-EFAULT);
167	}
168
169	tmp = src;
170	total = rq->inputsize;
171	while (total) {
172		unsigned int page_copycnt =
173			min_t(unsigned int, total, PAGE_SIZE - *inputmargin);
174
175		if (!inpage)
176			inpage = kmap_local_page(*in);
177		memcpy(tmp, inpage + *inputmargin, page_copycnt);
178		kunmap_local(inpage);
179		inpage = NULL;
180		tmp += page_copycnt;
181		total -= page_copycnt;
182		++in;
183		*inputmargin = 0;
184	}
185	*maptype = 2;
186	return src;
187}
188
189/*
190 * Get the exact inputsize with zero_padding feature.
191 *  - For LZ4, it should work if zero_padding feature is on (5.3+);
192 *  - For MicroLZMA, it'd be enabled all the time.
193 */
194int z_erofs_fixup_insize(struct z_erofs_decompress_req *rq, const char *padbuf,
195			 unsigned int padbufsize)
196{
197	const char *padend;
198
199	padend = memchr_inv(padbuf, 0, padbufsize);
200	if (!padend)
201		return -EFSCORRUPTED;
202	rq->inputsize -= padend - padbuf;
203	rq->pageofs_in += padend - padbuf;
204	return 0;
205}
206
207static int z_erofs_lz4_decompress_mem(struct z_erofs_lz4_decompress_ctx *ctx,
208				      u8 *dst)
209{
210	struct z_erofs_decompress_req *rq = ctx->rq;
211	bool support_0padding = false, may_inplace = false;
212	unsigned int inputmargin;
213	u8 *out, *headpage, *src;
214	int ret, maptype;
215
216	DBG_BUGON(*rq->in == NULL);
217	headpage = kmap_local_page(*rq->in);
218
219	/* LZ4 decompression inplace is only safe if zero_padding is enabled */
220	if (erofs_sb_has_zero_padding(EROFS_SB(rq->sb))) {
221		support_0padding = true;
222		ret = z_erofs_fixup_insize(rq, headpage + rq->pageofs_in,
223				min_t(unsigned int, rq->inputsize,
224				      rq->sb->s_blocksize - rq->pageofs_in));
225		if (ret) {
226			kunmap_local(headpage);
227			return ret;
228		}
229		may_inplace = !((rq->pageofs_in + rq->inputsize) &
230				(rq->sb->s_blocksize - 1));
231	}
232
233	inputmargin = rq->pageofs_in;
234	src = z_erofs_lz4_handle_overlap(ctx, headpage, dst, &inputmargin,
235					 &maptype, may_inplace);
236	if (IS_ERR(src))
237		return PTR_ERR(src);
238
239	out = dst + rq->pageofs_out;
240	/* legacy format could compress extra data in a pcluster. */
241	if (rq->partial_decoding || !support_0padding)
242		ret = LZ4_decompress_safe_partial(src + inputmargin, out,
243				rq->inputsize, rq->outputsize, rq->outputsize);
244	else
245		ret = LZ4_decompress_safe(src + inputmargin, out,
246					  rq->inputsize, rq->outputsize);
247
248	if (ret != rq->outputsize) {
249		erofs_err(rq->sb, "failed to decompress %d in[%u, %u] out[%u]",
250			  ret, rq->inputsize, inputmargin, rq->outputsize);
251		if (ret >= 0)
252			memset(out + ret, 0, rq->outputsize - ret);
253		ret = -EFSCORRUPTED;
254	} else {
255		ret = 0;
256	}
257
258	if (maptype == 0) {
259		kunmap_local(headpage);
260	} else if (maptype == 1) {
261		vm_unmap_ram(src, ctx->inpages);
262	} else if (maptype == 2) {
263		erofs_put_pcpubuf(src);
264	} else if (maptype != 3) {
265		DBG_BUGON(1);
266		return -EFAULT;
267	}
268	return ret;
269}
270
271static int z_erofs_lz4_decompress(struct z_erofs_decompress_req *rq,
272				  struct page **pagepool)
273{
274	struct z_erofs_lz4_decompress_ctx ctx;
275	unsigned int dst_maptype;
276	void *dst;
277	int ret;
278
279	ctx.rq = rq;
280	ctx.oend = rq->pageofs_out + rq->outputsize;
281	ctx.outpages = PAGE_ALIGN(ctx.oend) >> PAGE_SHIFT;
282	ctx.inpages = PAGE_ALIGN(rq->inputsize) >> PAGE_SHIFT;
283
284	/* one optimized fast path only for non bigpcluster cases yet */
285	if (ctx.inpages == 1 && ctx.outpages == 1 && !rq->inplace_io) {
286		DBG_BUGON(!*rq->out);
287		dst = kmap_local_page(*rq->out);
288		dst_maptype = 0;
289		goto dstmap_out;
290	}
291
292	/* general decoding path which can be used for all cases */
293	ret = z_erofs_lz4_prepare_dstpages(&ctx, pagepool);
294	if (ret < 0) {
295		return ret;
296	} else if (ret > 0) {
297		dst = page_address(*rq->out);
298		dst_maptype = 1;
299	} else {
300		dst = erofs_vm_map_ram(rq->out, ctx.outpages);
301		if (!dst)
302			return -ENOMEM;
303		dst_maptype = 2;
304	}
305
306dstmap_out:
307	ret = z_erofs_lz4_decompress_mem(&ctx, dst);
308	if (!dst_maptype)
309		kunmap_local(dst);
310	else if (dst_maptype == 2)
311		vm_unmap_ram(dst, ctx.outpages);
312	return ret;
313}
314
315static int z_erofs_transform_plain(struct z_erofs_decompress_req *rq,
316				   struct page **pagepool)
317{
318	const unsigned int nrpages_in =
319		PAGE_ALIGN(rq->pageofs_in + rq->inputsize) >> PAGE_SHIFT;
320	const unsigned int nrpages_out =
321		PAGE_ALIGN(rq->pageofs_out + rq->outputsize) >> PAGE_SHIFT;
322	const unsigned int bs = rq->sb->s_blocksize;
323	unsigned int cur = 0, ni = 0, no, pi, po, insz, cnt;
324	u8 *kin;
325
326	if (rq->outputsize > rq->inputsize)
327		return -EOPNOTSUPP;
328	if (rq->alg == Z_EROFS_COMPRESSION_INTERLACED) {
329		cur = bs - (rq->pageofs_out & (bs - 1));
330		pi = (rq->pageofs_in + rq->inputsize - cur) & ~PAGE_MASK;
331		cur = min(cur, rq->outputsize);
332		if (cur && rq->out[0]) {
333			kin = kmap_local_page(rq->in[nrpages_in - 1]);
334			if (rq->out[0] == rq->in[nrpages_in - 1]) {
335				memmove(kin + rq->pageofs_out, kin + pi, cur);
336				flush_dcache_page(rq->out[0]);
337			} else {
338				memcpy_to_page(rq->out[0], rq->pageofs_out,
339					       kin + pi, cur);
340			}
341			kunmap_local(kin);
342		}
343		rq->outputsize -= cur;
344	}
345
346	for (; rq->outputsize; rq->pageofs_in = 0, cur += PAGE_SIZE, ni++) {
347		insz = min(PAGE_SIZE - rq->pageofs_in, rq->outputsize);
348		rq->outputsize -= insz;
349		if (!rq->in[ni])
350			continue;
351		kin = kmap_local_page(rq->in[ni]);
352		pi = 0;
353		do {
354			no = (rq->pageofs_out + cur + pi) >> PAGE_SHIFT;
355			po = (rq->pageofs_out + cur + pi) & ~PAGE_MASK;
356			DBG_BUGON(no >= nrpages_out);
357			cnt = min(insz - pi, PAGE_SIZE - po);
358			if (rq->out[no] == rq->in[ni]) {
359				memmove(kin + po,
360					kin + rq->pageofs_in + pi, cnt);
361				flush_dcache_page(rq->out[no]);
362			} else if (rq->out[no]) {
363				memcpy_to_page(rq->out[no], po,
364					       kin + rq->pageofs_in + pi, cnt);
365			}
366			pi += cnt;
367		} while (pi < insz);
368		kunmap_local(kin);
369	}
370	DBG_BUGON(ni > nrpages_in);
371	return 0;
372}
373
374const struct z_erofs_decompressor erofs_decompressors[] = {
375	[Z_EROFS_COMPRESSION_SHIFTED] = {
376		.decompress = z_erofs_transform_plain,
377		.name = "shifted"
378	},
379	[Z_EROFS_COMPRESSION_INTERLACED] = {
380		.decompress = z_erofs_transform_plain,
381		.name = "interlaced"
382	},
383	[Z_EROFS_COMPRESSION_LZ4] = {
384		.config = z_erofs_load_lz4_config,
385		.decompress = z_erofs_lz4_decompress,
386		.name = "lz4"
387	},
388#ifdef CONFIG_EROFS_FS_ZIP_LZMA
389	[Z_EROFS_COMPRESSION_LZMA] = {
390		.config = z_erofs_load_lzma_config,
391		.decompress = z_erofs_lzma_decompress,
392		.name = "lzma"
393	},
394#endif
395#ifdef CONFIG_EROFS_FS_ZIP_DEFLATE
396	[Z_EROFS_COMPRESSION_DEFLATE] = {
397		.config = z_erofs_load_deflate_config,
398		.decompress = z_erofs_deflate_decompress,
399		.name = "deflate"
400	},
401#endif
402};
403
404int z_erofs_parse_cfgs(struct super_block *sb, struct erofs_super_block *dsb)
405{
406	struct erofs_sb_info *sbi = EROFS_SB(sb);
407	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
408	unsigned int algs, alg;
409	erofs_off_t offset;
410	int size, ret = 0;
411
412	if (!erofs_sb_has_compr_cfgs(sbi)) {
413		sbi->available_compr_algs = 1 << Z_EROFS_COMPRESSION_LZ4;
414		return z_erofs_load_lz4_config(sb, dsb, NULL, 0);
415	}
416
417	sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
418	if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
419		erofs_err(sb, "unidentified algorithms %x, please upgrade kernel",
420			  sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
421		return -EOPNOTSUPP;
422	}
423
424	erofs_init_metabuf(&buf, sb);
425	offset = EROFS_SUPER_OFFSET + sbi->sb_size;
426	alg = 0;
427	for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
428		void *data;
429
430		if (!(algs & 1))
431			continue;
432
433		data = erofs_read_metadata(sb, &buf, &offset, &size);
434		if (IS_ERR(data)) {
435			ret = PTR_ERR(data);
436			break;
437		}
438
439		if (alg >= ARRAY_SIZE(erofs_decompressors) ||
440		    !erofs_decompressors[alg].config) {
441			erofs_err(sb, "algorithm %d isn't enabled on this kernel",
442				  alg);
443			ret = -EOPNOTSUPP;
444		} else {
445			ret = erofs_decompressors[alg].config(sb,
446					dsb, data, size);
447		}
448
449		kfree(data);
450		if (ret)
451			break;
452	}
453	erofs_put_metabuf(&buf);
454	return ret;
455}
456