xref: /openbmc/linux/fs/erofs/zmap.c (revision a2cfe94a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2018-2019 HUAWEI, Inc.
4  *             https://www.huawei.com/
5  */
6 #include "internal.h"
7 #include <asm/unaligned.h>
8 #include <trace/events/erofs.h>
9 
10 struct z_erofs_maprecorder {
11 	struct inode *inode;
12 	struct erofs_map_blocks *map;
13 	void *kaddr;
14 
15 	unsigned long lcn;
16 	/* compression extent information gathered */
17 	u8  type, headtype;
18 	u16 clusterofs;
19 	u16 delta[2];
20 	erofs_blk_t pblk, compressedblks;
21 	erofs_off_t nextpackoff;
22 	bool partialref;
23 };
24 
z_erofs_load_full_lcluster(struct z_erofs_maprecorder * m,unsigned long lcn)25 static int z_erofs_load_full_lcluster(struct z_erofs_maprecorder *m,
26 				      unsigned long lcn)
27 {
28 	struct inode *const inode = m->inode;
29 	struct erofs_inode *const vi = EROFS_I(inode);
30 	const erofs_off_t pos = Z_EROFS_FULL_INDEX_ALIGN(erofs_iloc(inode) +
31 			vi->inode_isize + vi->xattr_isize) +
32 			lcn * sizeof(struct z_erofs_lcluster_index);
33 	struct z_erofs_lcluster_index *di;
34 	unsigned int advise, type;
35 
36 	m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
37 				      erofs_blknr(inode->i_sb, pos), EROFS_KMAP);
38 	if (IS_ERR(m->kaddr))
39 		return PTR_ERR(m->kaddr);
40 
41 	m->nextpackoff = pos + sizeof(struct z_erofs_lcluster_index);
42 	m->lcn = lcn;
43 	di = m->kaddr + erofs_blkoff(inode->i_sb, pos);
44 
45 	advise = le16_to_cpu(di->di_advise);
46 	type = (advise >> Z_EROFS_LI_LCLUSTER_TYPE_BIT) &
47 		((1 << Z_EROFS_LI_LCLUSTER_TYPE_BITS) - 1);
48 	switch (type) {
49 	case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
50 		m->clusterofs = 1 << vi->z_logical_clusterbits;
51 		m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
52 		if (m->delta[0] & Z_EROFS_LI_D0_CBLKCNT) {
53 			if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
54 					Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
55 				DBG_BUGON(1);
56 				return -EFSCORRUPTED;
57 			}
58 			m->compressedblks = m->delta[0] &
59 				~Z_EROFS_LI_D0_CBLKCNT;
60 			m->delta[0] = 1;
61 		}
62 		m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
63 		break;
64 	case Z_EROFS_LCLUSTER_TYPE_PLAIN:
65 	case Z_EROFS_LCLUSTER_TYPE_HEAD1:
66 	case Z_EROFS_LCLUSTER_TYPE_HEAD2:
67 		if (advise & Z_EROFS_LI_PARTIAL_REF)
68 			m->partialref = true;
69 		m->clusterofs = le16_to_cpu(di->di_clusterofs);
70 		if (m->clusterofs >= 1 << vi->z_logical_clusterbits) {
71 			DBG_BUGON(1);
72 			return -EFSCORRUPTED;
73 		}
74 		m->pblk = le32_to_cpu(di->di_u.blkaddr);
75 		break;
76 	default:
77 		DBG_BUGON(1);
78 		return -EOPNOTSUPP;
79 	}
80 	m->type = type;
81 	return 0;
82 }
83 
decode_compactedbits(unsigned int lobits,u8 * in,unsigned int pos,u8 * type)84 static unsigned int decode_compactedbits(unsigned int lobits,
85 					 u8 *in, unsigned int pos, u8 *type)
86 {
87 	const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
88 	const unsigned int lo = v & ((1 << lobits) - 1);
89 
90 	*type = (v >> lobits) & 3;
91 	return lo;
92 }
93 
get_compacted_la_distance(unsigned int lobits,unsigned int encodebits,unsigned int vcnt,u8 * in,int i)94 static int get_compacted_la_distance(unsigned int lobits,
95 				     unsigned int encodebits,
96 				     unsigned int vcnt, u8 *in, int i)
97 {
98 	unsigned int lo, d1 = 0;
99 	u8 type;
100 
101 	DBG_BUGON(i >= vcnt);
102 
103 	do {
104 		lo = decode_compactedbits(lobits, in, encodebits * i, &type);
105 
106 		if (type != Z_EROFS_LCLUSTER_TYPE_NONHEAD)
107 			return d1;
108 		++d1;
109 	} while (++i < vcnt);
110 
111 	/* vcnt - 1 (Z_EROFS_LCLUSTER_TYPE_NONHEAD) item */
112 	if (!(lo & Z_EROFS_LI_D0_CBLKCNT))
113 		d1 += lo - 1;
114 	return d1;
115 }
116 
unpack_compacted_index(struct z_erofs_maprecorder * m,unsigned int amortizedshift,erofs_off_t pos,bool lookahead)117 static int unpack_compacted_index(struct z_erofs_maprecorder *m,
118 				  unsigned int amortizedshift,
119 				  erofs_off_t pos, bool lookahead)
120 {
121 	struct erofs_inode *const vi = EROFS_I(m->inode);
122 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
123 	unsigned int vcnt, base, lo, lobits, encodebits, nblk, eofs;
124 	int i;
125 	u8 *in, type;
126 	bool big_pcluster;
127 
128 	if (1 << amortizedshift == 4 && lclusterbits <= 14)
129 		vcnt = 2;
130 	else if (1 << amortizedshift == 2 && lclusterbits <= 12)
131 		vcnt = 16;
132 	else
133 		return -EOPNOTSUPP;
134 
135 	/* it doesn't equal to round_up(..) */
136 	m->nextpackoff = round_down(pos, vcnt << amortizedshift) +
137 			 (vcnt << amortizedshift);
138 	big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1;
139 	lobits = max(lclusterbits, ilog2(Z_EROFS_LI_D0_CBLKCNT) + 1U);
140 	encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
141 	eofs = erofs_blkoff(m->inode->i_sb, pos);
142 	base = round_down(eofs, vcnt << amortizedshift);
143 	in = m->kaddr + base;
144 
145 	i = (eofs - base) >> amortizedshift;
146 
147 	lo = decode_compactedbits(lobits, in, encodebits * i, &type);
148 	m->type = type;
149 	if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
150 		m->clusterofs = 1 << lclusterbits;
151 
152 		/* figure out lookahead_distance: delta[1] if needed */
153 		if (lookahead)
154 			m->delta[1] = get_compacted_la_distance(lobits,
155 						encodebits, vcnt, in, i);
156 		if (lo & Z_EROFS_LI_D0_CBLKCNT) {
157 			if (!big_pcluster) {
158 				DBG_BUGON(1);
159 				return -EFSCORRUPTED;
160 			}
161 			m->compressedblks = lo & ~Z_EROFS_LI_D0_CBLKCNT;
162 			m->delta[0] = 1;
163 			return 0;
164 		} else if (i + 1 != (int)vcnt) {
165 			m->delta[0] = lo;
166 			return 0;
167 		}
168 		/*
169 		 * since the last lcluster in the pack is special,
170 		 * of which lo saves delta[1] rather than delta[0].
171 		 * Hence, get delta[0] by the previous lcluster indirectly.
172 		 */
173 		lo = decode_compactedbits(lobits, in,
174 					  encodebits * (i - 1), &type);
175 		if (type != Z_EROFS_LCLUSTER_TYPE_NONHEAD)
176 			lo = 0;
177 		else if (lo & Z_EROFS_LI_D0_CBLKCNT)
178 			lo = 1;
179 		m->delta[0] = lo + 1;
180 		return 0;
181 	}
182 	m->clusterofs = lo;
183 	m->delta[0] = 0;
184 	/* figout out blkaddr (pblk) for HEAD lclusters */
185 	if (!big_pcluster) {
186 		nblk = 1;
187 		while (i > 0) {
188 			--i;
189 			lo = decode_compactedbits(lobits, in,
190 						  encodebits * i, &type);
191 			if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD)
192 				i -= lo;
193 
194 			if (i >= 0)
195 				++nblk;
196 		}
197 	} else {
198 		nblk = 0;
199 		while (i > 0) {
200 			--i;
201 			lo = decode_compactedbits(lobits, in,
202 						  encodebits * i, &type);
203 			if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
204 				if (lo & Z_EROFS_LI_D0_CBLKCNT) {
205 					--i;
206 					nblk += lo & ~Z_EROFS_LI_D0_CBLKCNT;
207 					continue;
208 				}
209 				/* bigpcluster shouldn't have plain d0 == 1 */
210 				if (lo <= 1) {
211 					DBG_BUGON(1);
212 					return -EFSCORRUPTED;
213 				}
214 				i -= lo - 2;
215 				continue;
216 			}
217 			++nblk;
218 		}
219 	}
220 	in += (vcnt << amortizedshift) - sizeof(__le32);
221 	m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
222 	return 0;
223 }
224 
z_erofs_load_compact_lcluster(struct z_erofs_maprecorder * m,unsigned long lcn,bool lookahead)225 static int z_erofs_load_compact_lcluster(struct z_erofs_maprecorder *m,
226 					 unsigned long lcn, bool lookahead)
227 {
228 	struct inode *const inode = m->inode;
229 	struct erofs_inode *const vi = EROFS_I(inode);
230 	const erofs_off_t ebase = sizeof(struct z_erofs_map_header) +
231 		ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8);
232 	unsigned int totalidx = erofs_iblks(inode);
233 	unsigned int compacted_4b_initial, compacted_2b;
234 	unsigned int amortizedshift;
235 	erofs_off_t pos;
236 
237 	if (lcn >= totalidx)
238 		return -EINVAL;
239 
240 	m->lcn = lcn;
241 	/* used to align to 32-byte (compacted_2b) alignment */
242 	compacted_4b_initial = (32 - ebase % 32) / 4;
243 	if (compacted_4b_initial == 32 / 4)
244 		compacted_4b_initial = 0;
245 
246 	if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
247 	    compacted_4b_initial < totalidx)
248 		compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
249 	else
250 		compacted_2b = 0;
251 
252 	pos = ebase;
253 	if (lcn < compacted_4b_initial) {
254 		amortizedshift = 2;
255 		goto out;
256 	}
257 	pos += compacted_4b_initial * 4;
258 	lcn -= compacted_4b_initial;
259 
260 	if (lcn < compacted_2b) {
261 		amortizedshift = 1;
262 		goto out;
263 	}
264 	pos += compacted_2b * 2;
265 	lcn -= compacted_2b;
266 	amortizedshift = 2;
267 out:
268 	pos += lcn * (1 << amortizedshift);
269 	m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
270 				      erofs_blknr(inode->i_sb, pos), EROFS_KMAP);
271 	if (IS_ERR(m->kaddr))
272 		return PTR_ERR(m->kaddr);
273 	return unpack_compacted_index(m, amortizedshift, pos, lookahead);
274 }
275 
z_erofs_load_lcluster_from_disk(struct z_erofs_maprecorder * m,unsigned int lcn,bool lookahead)276 static int z_erofs_load_lcluster_from_disk(struct z_erofs_maprecorder *m,
277 					   unsigned int lcn, bool lookahead)
278 {
279 	switch (EROFS_I(m->inode)->datalayout) {
280 	case EROFS_INODE_COMPRESSED_FULL:
281 		return z_erofs_load_full_lcluster(m, lcn);
282 	case EROFS_INODE_COMPRESSED_COMPACT:
283 		return z_erofs_load_compact_lcluster(m, lcn, lookahead);
284 	default:
285 		return -EINVAL;
286 	}
287 }
288 
z_erofs_extent_lookback(struct z_erofs_maprecorder * m,unsigned int lookback_distance)289 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
290 				   unsigned int lookback_distance)
291 {
292 	struct super_block *sb = m->inode->i_sb;
293 	struct erofs_inode *const vi = EROFS_I(m->inode);
294 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
295 
296 	while (m->lcn >= lookback_distance) {
297 		unsigned long lcn = m->lcn - lookback_distance;
298 		int err;
299 
300 		err = z_erofs_load_lcluster_from_disk(m, lcn, false);
301 		if (err)
302 			return err;
303 
304 		switch (m->type) {
305 		case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
306 			lookback_distance = m->delta[0];
307 			if (!lookback_distance)
308 				goto err_bogus;
309 			continue;
310 		case Z_EROFS_LCLUSTER_TYPE_PLAIN:
311 		case Z_EROFS_LCLUSTER_TYPE_HEAD1:
312 		case Z_EROFS_LCLUSTER_TYPE_HEAD2:
313 			m->headtype = m->type;
314 			m->map->m_la = (lcn << lclusterbits) | m->clusterofs;
315 			return 0;
316 		default:
317 			erofs_err(sb, "unknown type %u @ lcn %lu of nid %llu",
318 				  m->type, lcn, vi->nid);
319 			DBG_BUGON(1);
320 			return -EOPNOTSUPP;
321 		}
322 	}
323 err_bogus:
324 	erofs_err(sb, "bogus lookback distance %u @ lcn %lu of nid %llu",
325 		  lookback_distance, m->lcn, vi->nid);
326 	DBG_BUGON(1);
327 	return -EFSCORRUPTED;
328 }
329 
z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder * m,unsigned int initial_lcn)330 static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m,
331 					    unsigned int initial_lcn)
332 {
333 	struct super_block *sb = m->inode->i_sb;
334 	struct erofs_inode *const vi = EROFS_I(m->inode);
335 	struct erofs_map_blocks *const map = m->map;
336 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
337 	unsigned long lcn;
338 	int err;
339 
340 	DBG_BUGON(m->type != Z_EROFS_LCLUSTER_TYPE_PLAIN &&
341 		  m->type != Z_EROFS_LCLUSTER_TYPE_HEAD1 &&
342 		  m->type != Z_EROFS_LCLUSTER_TYPE_HEAD2);
343 	DBG_BUGON(m->type != m->headtype);
344 
345 	if (m->headtype == Z_EROFS_LCLUSTER_TYPE_PLAIN ||
346 	    ((m->headtype == Z_EROFS_LCLUSTER_TYPE_HEAD1) &&
347 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1)) ||
348 	    ((m->headtype == Z_EROFS_LCLUSTER_TYPE_HEAD2) &&
349 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
350 		map->m_plen = 1ULL << lclusterbits;
351 		return 0;
352 	}
353 	lcn = m->lcn + 1;
354 	if (m->compressedblks)
355 		goto out;
356 
357 	err = z_erofs_load_lcluster_from_disk(m, lcn, false);
358 	if (err)
359 		return err;
360 
361 	/*
362 	 * If the 1st NONHEAD lcluster has already been handled initially w/o
363 	 * valid compressedblks, which means at least it mustn't be CBLKCNT, or
364 	 * an internal implemenatation error is detected.
365 	 *
366 	 * The following code can also handle it properly anyway, but let's
367 	 * BUG_ON in the debugging mode only for developers to notice that.
368 	 */
369 	DBG_BUGON(lcn == initial_lcn &&
370 		  m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD);
371 
372 	switch (m->type) {
373 	case Z_EROFS_LCLUSTER_TYPE_PLAIN:
374 	case Z_EROFS_LCLUSTER_TYPE_HEAD1:
375 	case Z_EROFS_LCLUSTER_TYPE_HEAD2:
376 		/*
377 		 * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type
378 		 * rather than CBLKCNT, it's a 1 lcluster-sized pcluster.
379 		 */
380 		m->compressedblks = 1 << (lclusterbits - sb->s_blocksize_bits);
381 		break;
382 	case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
383 		if (m->delta[0] != 1)
384 			goto err_bonus_cblkcnt;
385 		if (m->compressedblks)
386 			break;
387 		fallthrough;
388 	default:
389 		erofs_err(sb, "cannot found CBLKCNT @ lcn %lu of nid %llu", lcn,
390 			  vi->nid);
391 		DBG_BUGON(1);
392 		return -EFSCORRUPTED;
393 	}
394 out:
395 	map->m_plen = erofs_pos(sb, m->compressedblks);
396 	return 0;
397 err_bonus_cblkcnt:
398 	erofs_err(sb, "bogus CBLKCNT @ lcn %lu of nid %llu", lcn, vi->nid);
399 	DBG_BUGON(1);
400 	return -EFSCORRUPTED;
401 }
402 
z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder * m)403 static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m)
404 {
405 	struct inode *inode = m->inode;
406 	struct erofs_inode *vi = EROFS_I(inode);
407 	struct erofs_map_blocks *map = m->map;
408 	unsigned int lclusterbits = vi->z_logical_clusterbits;
409 	u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits;
410 	int err;
411 
412 	do {
413 		/* handle the last EOF pcluster (no next HEAD lcluster) */
414 		if ((lcn << lclusterbits) >= inode->i_size) {
415 			map->m_llen = inode->i_size - map->m_la;
416 			return 0;
417 		}
418 
419 		err = z_erofs_load_lcluster_from_disk(m, lcn, true);
420 		if (err)
421 			return err;
422 
423 		if (m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
424 			DBG_BUGON(!m->delta[1] &&
425 				  m->clusterofs != 1 << lclusterbits);
426 		} else if (m->type == Z_EROFS_LCLUSTER_TYPE_PLAIN ||
427 			   m->type == Z_EROFS_LCLUSTER_TYPE_HEAD1 ||
428 			   m->type == Z_EROFS_LCLUSTER_TYPE_HEAD2) {
429 			/* go on until the next HEAD lcluster */
430 			if (lcn != headlcn)
431 				break;
432 			m->delta[1] = 1;
433 		} else {
434 			erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu",
435 				  m->type, lcn, vi->nid);
436 			DBG_BUGON(1);
437 			return -EOPNOTSUPP;
438 		}
439 		lcn += m->delta[1];
440 	} while (m->delta[1]);
441 
442 	map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la;
443 	return 0;
444 }
445 
z_erofs_do_map_blocks(struct inode * inode,struct erofs_map_blocks * map,int flags)446 static int z_erofs_do_map_blocks(struct inode *inode,
447 				 struct erofs_map_blocks *map, int flags)
448 {
449 	struct erofs_inode *const vi = EROFS_I(inode);
450 	bool ztailpacking = vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER;
451 	bool fragment = vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
452 	struct z_erofs_maprecorder m = {
453 		.inode = inode,
454 		.map = map,
455 	};
456 	int err = 0;
457 	unsigned int lclusterbits, endoff, afmt;
458 	unsigned long initial_lcn;
459 	unsigned long long ofs, end;
460 
461 	lclusterbits = vi->z_logical_clusterbits;
462 	ofs = flags & EROFS_GET_BLOCKS_FINDTAIL ? inode->i_size - 1 : map->m_la;
463 	initial_lcn = ofs >> lclusterbits;
464 	endoff = ofs & ((1 << lclusterbits) - 1);
465 
466 	err = z_erofs_load_lcluster_from_disk(&m, initial_lcn, false);
467 	if (err)
468 		goto unmap_out;
469 
470 	if (ztailpacking && (flags & EROFS_GET_BLOCKS_FINDTAIL))
471 		vi->z_idataoff = m.nextpackoff;
472 
473 	map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED;
474 	end = (m.lcn + 1ULL) << lclusterbits;
475 
476 	switch (m.type) {
477 	case Z_EROFS_LCLUSTER_TYPE_PLAIN:
478 	case Z_EROFS_LCLUSTER_TYPE_HEAD1:
479 	case Z_EROFS_LCLUSTER_TYPE_HEAD2:
480 		if (endoff >= m.clusterofs) {
481 			m.headtype = m.type;
482 			map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
483 			/*
484 			 * For ztailpacking files, in order to inline data more
485 			 * effectively, special EOF lclusters are now supported
486 			 * which can have three parts at most.
487 			 */
488 			if (ztailpacking && end > inode->i_size)
489 				end = inode->i_size;
490 			break;
491 		}
492 		/* m.lcn should be >= 1 if endoff < m.clusterofs */
493 		if (!m.lcn) {
494 			erofs_err(inode->i_sb,
495 				  "invalid logical cluster 0 at nid %llu",
496 				  vi->nid);
497 			err = -EFSCORRUPTED;
498 			goto unmap_out;
499 		}
500 		end = (m.lcn << lclusterbits) | m.clusterofs;
501 		map->m_flags |= EROFS_MAP_FULL_MAPPED;
502 		m.delta[0] = 1;
503 		fallthrough;
504 	case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
505 		/* get the corresponding first chunk */
506 		err = z_erofs_extent_lookback(&m, m.delta[0]);
507 		if (err)
508 			goto unmap_out;
509 		break;
510 	default:
511 		erofs_err(inode->i_sb,
512 			  "unknown type %u @ offset %llu of nid %llu",
513 			  m.type, ofs, vi->nid);
514 		err = -EOPNOTSUPP;
515 		goto unmap_out;
516 	}
517 	if (m.partialref)
518 		map->m_flags |= EROFS_MAP_PARTIAL_REF;
519 	map->m_llen = end - map->m_la;
520 
521 	if (flags & EROFS_GET_BLOCKS_FINDTAIL) {
522 		vi->z_tailextent_headlcn = m.lcn;
523 		/* for non-compact indexes, fragmentoff is 64 bits */
524 		if (fragment && vi->datalayout == EROFS_INODE_COMPRESSED_FULL)
525 			vi->z_fragmentoff |= (u64)m.pblk << 32;
526 	}
527 	if (ztailpacking && m.lcn == vi->z_tailextent_headlcn) {
528 		map->m_flags |= EROFS_MAP_META;
529 		map->m_pa = vi->z_idataoff;
530 		map->m_plen = vi->z_idata_size;
531 	} else if (fragment && m.lcn == vi->z_tailextent_headlcn) {
532 		map->m_flags |= EROFS_MAP_FRAGMENT;
533 	} else {
534 		map->m_pa = erofs_pos(inode->i_sb, m.pblk);
535 		err = z_erofs_get_extent_compressedlen(&m, initial_lcn);
536 		if (err)
537 			goto unmap_out;
538 	}
539 
540 	if (m.headtype == Z_EROFS_LCLUSTER_TYPE_PLAIN) {
541 		if (map->m_llen > map->m_plen) {
542 			DBG_BUGON(1);
543 			err = -EFSCORRUPTED;
544 			goto unmap_out;
545 		}
546 		afmt = vi->z_advise & Z_EROFS_ADVISE_INTERLACED_PCLUSTER ?
547 			Z_EROFS_COMPRESSION_INTERLACED :
548 			Z_EROFS_COMPRESSION_SHIFTED;
549 	} else {
550 		afmt = m.headtype == Z_EROFS_LCLUSTER_TYPE_HEAD2 ?
551 			vi->z_algorithmtype[1] : vi->z_algorithmtype[0];
552 		if (!(EROFS_I_SB(inode)->available_compr_algs & (1 << afmt))) {
553 			erofs_err(inode->i_sb, "inconsistent algorithmtype %u for nid %llu",
554 				  afmt, vi->nid);
555 			err = -EFSCORRUPTED;
556 			goto unmap_out;
557 		}
558 	}
559 	map->m_algorithmformat = afmt;
560 
561 	if ((flags & EROFS_GET_BLOCKS_FIEMAP) ||
562 	    ((flags & EROFS_GET_BLOCKS_READMORE) &&
563 	     (map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA ||
564 	      map->m_algorithmformat == Z_EROFS_COMPRESSION_DEFLATE) &&
565 	      map->m_llen >= i_blocksize(inode))) {
566 		err = z_erofs_get_extent_decompressedlen(&m);
567 		if (!err)
568 			map->m_flags |= EROFS_MAP_FULL_MAPPED;
569 	}
570 
571 unmap_out:
572 	erofs_unmap_metabuf(&m.map->buf);
573 	return err;
574 }
575 
z_erofs_fill_inode_lazy(struct inode * inode)576 static int z_erofs_fill_inode_lazy(struct inode *inode)
577 {
578 	struct erofs_inode *const vi = EROFS_I(inode);
579 	struct super_block *const sb = inode->i_sb;
580 	int err, headnr;
581 	erofs_off_t pos;
582 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
583 	void *kaddr;
584 	struct z_erofs_map_header *h;
585 
586 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) {
587 		/*
588 		 * paired with smp_mb() at the end of the function to ensure
589 		 * fields will only be observed after the bit is set.
590 		 */
591 		smp_mb();
592 		return 0;
593 	}
594 
595 	if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE))
596 		return -ERESTARTSYS;
597 
598 	err = 0;
599 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
600 		goto out_unlock;
601 
602 	pos = ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8);
603 	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(sb, pos), EROFS_KMAP);
604 	if (IS_ERR(kaddr)) {
605 		err = PTR_ERR(kaddr);
606 		goto out_unlock;
607 	}
608 
609 	h = kaddr + erofs_blkoff(sb, pos);
610 	/*
611 	 * if the highest bit of the 8-byte map header is set, the whole file
612 	 * is stored in the packed inode. The rest bits keeps z_fragmentoff.
613 	 */
614 	if (h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT) {
615 		vi->z_advise = Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
616 		vi->z_fragmentoff = le64_to_cpu(*(__le64 *)h) ^ (1ULL << 63);
617 		vi->z_tailextent_headlcn = 0;
618 		goto done;
619 	}
620 	vi->z_advise = le16_to_cpu(h->h_advise);
621 	vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
622 	vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
623 
624 	headnr = 0;
625 	if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX ||
626 	    vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) {
627 		erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel",
628 			  headnr + 1, vi->z_algorithmtype[headnr], vi->nid);
629 		err = -EOPNOTSUPP;
630 		goto out_put_metabuf;
631 	}
632 
633 	vi->z_logical_clusterbits = sb->s_blocksize_bits + (h->h_clusterbits & 7);
634 	if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
635 	    vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
636 			    Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
637 		erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu",
638 			  vi->nid);
639 		err = -EFSCORRUPTED;
640 		goto out_put_metabuf;
641 	}
642 	if (vi->datalayout == EROFS_INODE_COMPRESSED_COMPACT &&
643 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^
644 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
645 		erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu",
646 			  vi->nid);
647 		err = -EFSCORRUPTED;
648 		goto out_put_metabuf;
649 	}
650 
651 	if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
652 		struct erofs_map_blocks map = {
653 			.buf = __EROFS_BUF_INITIALIZER
654 		};
655 
656 		vi->z_idata_size = le16_to_cpu(h->h_idata_size);
657 		err = z_erofs_do_map_blocks(inode, &map,
658 					    EROFS_GET_BLOCKS_FINDTAIL);
659 		erofs_put_metabuf(&map.buf);
660 
661 		if (!map.m_plen ||
662 		    erofs_blkoff(sb, map.m_pa) + map.m_plen > sb->s_blocksize) {
663 			erofs_err(sb, "invalid tail-packing pclustersize %llu",
664 				  map.m_plen);
665 			err = -EFSCORRUPTED;
666 		}
667 		if (err < 0)
668 			goto out_put_metabuf;
669 	}
670 
671 	if (vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER &&
672 	    !(h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT)) {
673 		struct erofs_map_blocks map = {
674 			.buf = __EROFS_BUF_INITIALIZER
675 		};
676 
677 		vi->z_fragmentoff = le32_to_cpu(h->h_fragmentoff);
678 		err = z_erofs_do_map_blocks(inode, &map,
679 					    EROFS_GET_BLOCKS_FINDTAIL);
680 		erofs_put_metabuf(&map.buf);
681 		if (err < 0)
682 			goto out_put_metabuf;
683 	}
684 done:
685 	/* paired with smp_mb() at the beginning of the function */
686 	smp_mb();
687 	set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
688 out_put_metabuf:
689 	erofs_put_metabuf(&buf);
690 out_unlock:
691 	clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
692 	return err;
693 }
694 
z_erofs_map_blocks_iter(struct inode * inode,struct erofs_map_blocks * map,int flags)695 int z_erofs_map_blocks_iter(struct inode *inode, struct erofs_map_blocks *map,
696 			    int flags)
697 {
698 	struct erofs_inode *const vi = EROFS_I(inode);
699 	int err = 0;
700 
701 	trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
702 
703 	/* when trying to read beyond EOF, leave it unmapped */
704 	if (map->m_la >= inode->i_size) {
705 		map->m_llen = map->m_la + 1 - inode->i_size;
706 		map->m_la = inode->i_size;
707 		map->m_flags = 0;
708 		goto out;
709 	}
710 
711 	err = z_erofs_fill_inode_lazy(inode);
712 	if (err)
713 		goto out;
714 
715 	if ((vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER) &&
716 	    !vi->z_tailextent_headlcn) {
717 		map->m_la = 0;
718 		map->m_llen = inode->i_size;
719 		map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_FULL_MAPPED |
720 				EROFS_MAP_FRAGMENT;
721 		goto out;
722 	}
723 
724 	err = z_erofs_do_map_blocks(inode, map, flags);
725 out:
726 	if (err)
727 		map->m_llen = 0;
728 	trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
729 	return err;
730 }
731 
z_erofs_iomap_begin_report(struct inode * inode,loff_t offset,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)732 static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
733 				loff_t length, unsigned int flags,
734 				struct iomap *iomap, struct iomap *srcmap)
735 {
736 	int ret;
737 	struct erofs_map_blocks map = { .m_la = offset };
738 
739 	ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
740 	erofs_put_metabuf(&map.buf);
741 	if (ret < 0)
742 		return ret;
743 
744 	iomap->bdev = inode->i_sb->s_bdev;
745 	iomap->offset = map.m_la;
746 	iomap->length = map.m_llen;
747 	if (map.m_flags & EROFS_MAP_MAPPED) {
748 		iomap->type = IOMAP_MAPPED;
749 		iomap->addr = map.m_flags & EROFS_MAP_FRAGMENT ?
750 			      IOMAP_NULL_ADDR : map.m_pa;
751 	} else {
752 		iomap->type = IOMAP_HOLE;
753 		iomap->addr = IOMAP_NULL_ADDR;
754 		/*
755 		 * No strict rule on how to describe extents for post EOF, yet
756 		 * we need to do like below. Otherwise, iomap itself will get
757 		 * into an endless loop on post EOF.
758 		 *
759 		 * Calculate the effective offset by subtracting extent start
760 		 * (map.m_la) from the requested offset, and add it to length.
761 		 * (NB: offset >= map.m_la always)
762 		 */
763 		if (iomap->offset >= inode->i_size)
764 			iomap->length = length + offset - map.m_la;
765 	}
766 	iomap->flags = 0;
767 	return 0;
768 }
769 
770 const struct iomap_ops z_erofs_iomap_report_ops = {
771 	.iomap_begin = z_erofs_iomap_begin_report,
772 };
773