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/module.h>
8 #include <linux/lz4.h>
9
10 #ifndef LZ4_DISTANCE_MAX /* history window size */
11 #define LZ4_DISTANCE_MAX 65535 /* set to maximum value by default */
12 #endif
13
14 #define LZ4_MAX_DISTANCE_PAGES (DIV_ROUND_UP(LZ4_DISTANCE_MAX, PAGE_SIZE) + 1)
15 #ifndef LZ4_DECOMPRESS_INPLACE_MARGIN
16 #define LZ4_DECOMPRESS_INPLACE_MARGIN(srcsize) (((srcsize) >> 8) + 32)
17 #endif
18
19 struct z_erofs_lz4_decompress_ctx {
20 struct z_erofs_decompress_req *rq;
21 /* # of encoded, decoded pages */
22 unsigned int inpages, outpages;
23 /* decoded block total length (used for in-place decompression) */
24 unsigned int oend;
25 };
26
z_erofs_load_lz4_config(struct super_block * sb,struct erofs_super_block * dsb,void * data,int size)27 static int z_erofs_load_lz4_config(struct super_block *sb,
28 struct erofs_super_block *dsb, void *data, int size)
29 {
30 struct erofs_sb_info *sbi = EROFS_SB(sb);
31 struct z_erofs_lz4_cfgs *lz4 = data;
32 u16 distance;
33
34 if (lz4) {
35 if (size < sizeof(struct z_erofs_lz4_cfgs)) {
36 erofs_err(sb, "invalid lz4 cfgs, size=%u", size);
37 return -EINVAL;
38 }
39 distance = le16_to_cpu(lz4->max_distance);
40
41 sbi->lz4.max_pclusterblks = le16_to_cpu(lz4->max_pclusterblks);
42 if (!sbi->lz4.max_pclusterblks) {
43 sbi->lz4.max_pclusterblks = 1; /* reserved case */
44 } else if (sbi->lz4.max_pclusterblks >
45 erofs_blknr(sb, Z_EROFS_PCLUSTER_MAX_SIZE)) {
46 erofs_err(sb, "too large lz4 pclusterblks %u",
47 sbi->lz4.max_pclusterblks);
48 return -EINVAL;
49 }
50 } else {
51 distance = le16_to_cpu(dsb->u1.lz4_max_distance);
52 sbi->lz4.max_pclusterblks = 1;
53 }
54
55 sbi->lz4.max_distance_pages = distance ?
56 DIV_ROUND_UP(distance, PAGE_SIZE) + 1 :
57 LZ4_MAX_DISTANCE_PAGES;
58 return erofs_pcpubuf_growsize(sbi->lz4.max_pclusterblks);
59 }
60
61 /*
62 * Fill all gaps with bounce pages if it's a sparse page list. Also check if
63 * all physical pages are consecutive, which can be seen for moderate CR.
64 */
z_erofs_lz4_prepare_dstpages(struct z_erofs_lz4_decompress_ctx * ctx,struct page ** pagepool)65 static int z_erofs_lz4_prepare_dstpages(struct z_erofs_lz4_decompress_ctx *ctx,
66 struct page **pagepool)
67 {
68 struct z_erofs_decompress_req *rq = ctx->rq;
69 struct page *availables[LZ4_MAX_DISTANCE_PAGES] = { NULL };
70 unsigned long bounced[DIV_ROUND_UP(LZ4_MAX_DISTANCE_PAGES,
71 BITS_PER_LONG)] = { 0 };
72 unsigned int lz4_max_distance_pages =
73 EROFS_SB(rq->sb)->lz4.max_distance_pages;
74 void *kaddr = NULL;
75 unsigned int i, j, top;
76
77 top = 0;
78 for (i = j = 0; i < ctx->outpages; ++i, ++j) {
79 struct page *const page = rq->out[i];
80 struct page *victim;
81
82 if (j >= lz4_max_distance_pages)
83 j = 0;
84
85 /* 'valid' bounced can only be tested after a complete round */
86 if (!rq->fillgaps && test_bit(j, bounced)) {
87 DBG_BUGON(i < lz4_max_distance_pages);
88 DBG_BUGON(top >= lz4_max_distance_pages);
89 availables[top++] = rq->out[i - lz4_max_distance_pages];
90 }
91
92 if (page) {
93 __clear_bit(j, bounced);
94 if (!PageHighMem(page)) {
95 if (!i) {
96 kaddr = page_address(page);
97 continue;
98 }
99 if (kaddr &&
100 kaddr + PAGE_SIZE == page_address(page)) {
101 kaddr += PAGE_SIZE;
102 continue;
103 }
104 }
105 kaddr = NULL;
106 continue;
107 }
108 kaddr = NULL;
109 __set_bit(j, bounced);
110
111 if (top) {
112 victim = availables[--top];
113 get_page(victim);
114 } else {
115 victim = erofs_allocpage(pagepool,
116 GFP_KERNEL | __GFP_NOFAIL);
117 set_page_private(victim, Z_EROFS_SHORTLIVED_PAGE);
118 }
119 rq->out[i] = victim;
120 }
121 return kaddr ? 1 : 0;
122 }
123
z_erofs_lz4_handle_overlap(struct z_erofs_lz4_decompress_ctx * ctx,void * inpage,void * out,unsigned int * inputmargin,int * maptype,bool may_inplace)124 static 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
159 docopy:
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 */
z_erofs_fixup_insize(struct z_erofs_decompress_req * rq,const char * padbuf,unsigned int padbufsize)194 int 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
z_erofs_lz4_decompress_mem(struct z_erofs_lz4_decompress_ctx * ctx,u8 * dst)207 static 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
z_erofs_lz4_decompress(struct z_erofs_decompress_req * rq,struct page ** pagepool)271 static 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
306 dstmap_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
z_erofs_transform_plain(struct z_erofs_decompress_req * rq,struct page ** pagepool)315 static int z_erofs_transform_plain(struct z_erofs_decompress_req *rq,
316 struct page **pagepool)
317 {
318 const unsigned int inpages = PAGE_ALIGN(rq->inputsize) >> PAGE_SHIFT;
319 const unsigned int outpages =
320 PAGE_ALIGN(rq->pageofs_out + rq->outputsize) >> PAGE_SHIFT;
321 const unsigned int righthalf = min_t(unsigned int, rq->outputsize,
322 PAGE_SIZE - rq->pageofs_out);
323 const unsigned int lefthalf = rq->outputsize - righthalf;
324 const unsigned int interlaced_offset =
325 rq->alg == Z_EROFS_COMPRESSION_SHIFTED ? 0 : rq->pageofs_out;
326 u8 *src;
327
328 if (outpages > 2 && rq->alg == Z_EROFS_COMPRESSION_SHIFTED) {
329 DBG_BUGON(1);
330 return -EFSCORRUPTED;
331 }
332
333 if (rq->out[0] == *rq->in) {
334 DBG_BUGON(rq->pageofs_out);
335 return 0;
336 }
337
338 src = kmap_local_page(rq->in[inpages - 1]) + rq->pageofs_in;
339 if (rq->out[0])
340 memcpy_to_page(rq->out[0], rq->pageofs_out,
341 src + interlaced_offset, righthalf);
342
343 if (outpages > inpages) {
344 DBG_BUGON(!rq->out[outpages - 1]);
345 if (rq->out[outpages - 1] != rq->in[inpages - 1])
346 memcpy_to_page(rq->out[outpages - 1], 0, src +
347 (interlaced_offset ? 0 : righthalf),
348 lefthalf);
349 else if (!interlaced_offset)
350 memmove(src, src + righthalf, lefthalf);
351 }
352 kunmap_local(src);
353 return 0;
354 }
355
356 const struct z_erofs_decompressor erofs_decompressors[] = {
357 [Z_EROFS_COMPRESSION_SHIFTED] = {
358 .decompress = z_erofs_transform_plain,
359 .name = "shifted"
360 },
361 [Z_EROFS_COMPRESSION_INTERLACED] = {
362 .decompress = z_erofs_transform_plain,
363 .name = "interlaced"
364 },
365 [Z_EROFS_COMPRESSION_LZ4] = {
366 .config = z_erofs_load_lz4_config,
367 .decompress = z_erofs_lz4_decompress,
368 .name = "lz4"
369 },
370 #ifdef CONFIG_EROFS_FS_ZIP_LZMA
371 [Z_EROFS_COMPRESSION_LZMA] = {
372 .config = z_erofs_load_lzma_config,
373 .decompress = z_erofs_lzma_decompress,
374 .name = "lzma"
375 },
376 #endif
377 #ifdef CONFIG_EROFS_FS_ZIP_DEFLATE
378 [Z_EROFS_COMPRESSION_DEFLATE] = {
379 .config = z_erofs_load_deflate_config,
380 .decompress = z_erofs_deflate_decompress,
381 .name = "deflate"
382 },
383 #endif
384 };
385
z_erofs_parse_cfgs(struct super_block * sb,struct erofs_super_block * dsb)386 int z_erofs_parse_cfgs(struct super_block *sb, struct erofs_super_block *dsb)
387 {
388 struct erofs_sb_info *sbi = EROFS_SB(sb);
389 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
390 unsigned int algs, alg;
391 erofs_off_t offset;
392 int size, ret = 0;
393
394 if (!erofs_sb_has_compr_cfgs(sbi)) {
395 sbi->available_compr_algs = 1 << Z_EROFS_COMPRESSION_LZ4;
396 return z_erofs_load_lz4_config(sb, dsb, NULL, 0);
397 }
398
399 sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
400 if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
401 erofs_err(sb, "unidentified algorithms %x, please upgrade kernel",
402 sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
403 return -EOPNOTSUPP;
404 }
405
406 erofs_init_metabuf(&buf, sb);
407 offset = EROFS_SUPER_OFFSET + sbi->sb_size;
408 alg = 0;
409 for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
410 void *data;
411
412 if (!(algs & 1))
413 continue;
414
415 data = erofs_read_metadata(sb, &buf, &offset, &size);
416 if (IS_ERR(data)) {
417 ret = PTR_ERR(data);
418 break;
419 }
420
421 if (alg >= ARRAY_SIZE(erofs_decompressors) ||
422 !erofs_decompressors[alg].config) {
423 erofs_err(sb, "algorithm %d isn't enabled on this kernel",
424 alg);
425 ret = -EOPNOTSUPP;
426 } else {
427 ret = erofs_decompressors[alg].config(sb,
428 dsb, data, size);
429 }
430
431 kfree(data);
432 if (ret)
433 break;
434 }
435 erofs_put_metabuf(&buf);
436 return ret;
437 }
438