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
252 print_hex_dump(KERN_DEBUG, "[ in]: ", DUMP_PREFIX_OFFSET,
253 16, 1, src + inputmargin, rq->inputsize, true);
254 print_hex_dump(KERN_DEBUG, "[out]: ", DUMP_PREFIX_OFFSET,
255 16, 1, out, rq->outputsize, true);
256
257 if (ret >= 0)
258 memset(out + ret, 0, rq->outputsize - ret);
259 ret = -EIO;
260 } else {
261 ret = 0;
262 }
263
264 if (maptype == 0) {
265 kunmap_local(headpage);
266 } else if (maptype == 1) {
267 vm_unmap_ram(src, ctx->inpages);
268 } else if (maptype == 2) {
269 erofs_put_pcpubuf(src);
270 } else if (maptype != 3) {
271 DBG_BUGON(1);
272 return -EFAULT;
273 }
274 return ret;
275 }
276
z_erofs_lz4_decompress(struct z_erofs_decompress_req * rq,struct page ** pagepool)277 static int z_erofs_lz4_decompress(struct z_erofs_decompress_req *rq,
278 struct page **pagepool)
279 {
280 struct z_erofs_lz4_decompress_ctx ctx;
281 unsigned int dst_maptype;
282 void *dst;
283 int ret;
284
285 ctx.rq = rq;
286 ctx.oend = rq->pageofs_out + rq->outputsize;
287 ctx.outpages = PAGE_ALIGN(ctx.oend) >> PAGE_SHIFT;
288 ctx.inpages = PAGE_ALIGN(rq->inputsize) >> PAGE_SHIFT;
289
290 /* one optimized fast path only for non bigpcluster cases yet */
291 if (ctx.inpages == 1 && ctx.outpages == 1 && !rq->inplace_io) {
292 DBG_BUGON(!*rq->out);
293 dst = kmap_local_page(*rq->out);
294 dst_maptype = 0;
295 goto dstmap_out;
296 }
297
298 /* general decoding path which can be used for all cases */
299 ret = z_erofs_lz4_prepare_dstpages(&ctx, pagepool);
300 if (ret < 0) {
301 return ret;
302 } else if (ret > 0) {
303 dst = page_address(*rq->out);
304 dst_maptype = 1;
305 } else {
306 dst = erofs_vm_map_ram(rq->out, ctx.outpages);
307 if (!dst)
308 return -ENOMEM;
309 dst_maptype = 2;
310 }
311
312 dstmap_out:
313 ret = z_erofs_lz4_decompress_mem(&ctx, dst);
314 if (!dst_maptype)
315 kunmap_local(dst);
316 else if (dst_maptype == 2)
317 vm_unmap_ram(dst, ctx.outpages);
318 return ret;
319 }
320
z_erofs_transform_plain(struct z_erofs_decompress_req * rq,struct page ** pagepool)321 static int z_erofs_transform_plain(struct z_erofs_decompress_req *rq,
322 struct page **pagepool)
323 {
324 const unsigned int nrpages_in =
325 PAGE_ALIGN(rq->pageofs_in + rq->inputsize) >> PAGE_SHIFT;
326 const unsigned int nrpages_out =
327 PAGE_ALIGN(rq->pageofs_out + rq->outputsize) >> PAGE_SHIFT;
328 const unsigned int bs = rq->sb->s_blocksize;
329 unsigned int cur = 0, ni = 0, no, pi, po, insz, cnt;
330 u8 *kin;
331
332 DBG_BUGON(rq->outputsize > rq->inputsize);
333 if (rq->alg == Z_EROFS_COMPRESSION_INTERLACED) {
334 cur = bs - (rq->pageofs_out & (bs - 1));
335 pi = (rq->pageofs_in + rq->inputsize - cur) & ~PAGE_MASK;
336 cur = min(cur, rq->outputsize);
337 if (cur && rq->out[0]) {
338 kin = kmap_local_page(rq->in[nrpages_in - 1]);
339 if (rq->out[0] == rq->in[nrpages_in - 1]) {
340 memmove(kin + rq->pageofs_out, kin + pi, cur);
341 flush_dcache_page(rq->out[0]);
342 } else {
343 memcpy_to_page(rq->out[0], rq->pageofs_out,
344 kin + pi, cur);
345 }
346 kunmap_local(kin);
347 }
348 rq->outputsize -= cur;
349 }
350
351 for (; rq->outputsize; rq->pageofs_in = 0, cur += PAGE_SIZE, ni++) {
352 insz = min_t(unsigned int, PAGE_SIZE - rq->pageofs_in,
353 rq->outputsize);
354 rq->outputsize -= insz;
355 if (!rq->in[ni])
356 continue;
357 kin = kmap_local_page(rq->in[ni]);
358 pi = 0;
359 do {
360 no = (rq->pageofs_out + cur + pi) >> PAGE_SHIFT;
361 po = (rq->pageofs_out + cur + pi) & ~PAGE_MASK;
362 DBG_BUGON(no >= nrpages_out);
363 cnt = min_t(unsigned int, insz - pi, PAGE_SIZE - po);
364 if (rq->out[no] == rq->in[ni]) {
365 memmove(kin + po,
366 kin + rq->pageofs_in + pi, cnt);
367 flush_dcache_page(rq->out[no]);
368 } else if (rq->out[no]) {
369 memcpy_to_page(rq->out[no], po,
370 kin + rq->pageofs_in + pi, cnt);
371 }
372 pi += cnt;
373 } while (pi < insz);
374 kunmap_local(kin);
375 }
376 DBG_BUGON(ni > nrpages_in);
377 return 0;
378 }
379
380 const struct z_erofs_decompressor erofs_decompressors[] = {
381 [Z_EROFS_COMPRESSION_SHIFTED] = {
382 .decompress = z_erofs_transform_plain,
383 .name = "shifted"
384 },
385 [Z_EROFS_COMPRESSION_INTERLACED] = {
386 .decompress = z_erofs_transform_plain,
387 .name = "interlaced"
388 },
389 [Z_EROFS_COMPRESSION_LZ4] = {
390 .config = z_erofs_load_lz4_config,
391 .decompress = z_erofs_lz4_decompress,
392 .name = "lz4"
393 },
394 #ifdef CONFIG_EROFS_FS_ZIP_LZMA
395 [Z_EROFS_COMPRESSION_LZMA] = {
396 .config = z_erofs_load_lzma_config,
397 .decompress = z_erofs_lzma_decompress,
398 .name = "lzma"
399 },
400 #endif
401 };
402
z_erofs_parse_cfgs(struct super_block * sb,struct erofs_super_block * dsb)403 int z_erofs_parse_cfgs(struct super_block *sb, struct erofs_super_block *dsb)
404 {
405 struct erofs_sb_info *sbi = EROFS_SB(sb);
406 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
407 unsigned int algs, alg;
408 erofs_off_t offset;
409 int size, ret = 0;
410
411 if (!erofs_sb_has_compr_cfgs(sbi)) {
412 sbi->available_compr_algs = 1 << Z_EROFS_COMPRESSION_LZ4;
413 return z_erofs_load_lz4_config(sb, dsb, NULL, 0);
414 }
415
416 sbi->available_compr_algs = le16_to_cpu(dsb->u1.available_compr_algs);
417 if (sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS) {
418 erofs_err(sb, "unidentified algorithms %x, please upgrade kernel",
419 sbi->available_compr_algs & ~Z_EROFS_ALL_COMPR_ALGS);
420 return -EOPNOTSUPP;
421 }
422
423 offset = EROFS_SUPER_OFFSET + sbi->sb_size;
424 alg = 0;
425 for (algs = sbi->available_compr_algs; algs; algs >>= 1, ++alg) {
426 void *data;
427
428 if (!(algs & 1))
429 continue;
430
431 data = erofs_read_metadata(sb, &buf, &offset, &size);
432 if (IS_ERR(data)) {
433 ret = PTR_ERR(data);
434 break;
435 }
436
437 if (alg >= ARRAY_SIZE(erofs_decompressors) ||
438 !erofs_decompressors[alg].config) {
439 erofs_err(sb, "algorithm %d isn't enabled on this kernel",
440 alg);
441 ret = -EOPNOTSUPP;
442 } else {
443 ret = erofs_decompressors[alg].config(sb,
444 dsb, data, size);
445 }
446
447 kfree(data);
448 if (ret)
449 break;
450 }
451 erofs_put_metabuf(&buf);
452 return ret;
453 }
454