xref: /openbmc/linux/fs/erofs/zmap.c (revision 5e0266f0)
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 int z_erofs_fill_inode(struct inode *inode)
11 {
12 	struct erofs_inode *const vi = EROFS_I(inode);
13 	struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
14 
15 	if (!erofs_sb_has_big_pcluster(sbi) &&
16 	    !erofs_sb_has_ztailpacking(sbi) && !erofs_sb_has_fragments(sbi) &&
17 	    vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY) {
18 		vi->z_advise = 0;
19 		vi->z_algorithmtype[0] = 0;
20 		vi->z_algorithmtype[1] = 0;
21 		vi->z_logical_clusterbits = LOG_BLOCK_SIZE;
22 		set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
23 	}
24 	inode->i_mapping->a_ops = &z_erofs_aops;
25 	return 0;
26 }
27 
28 struct z_erofs_maprecorder {
29 	struct inode *inode;
30 	struct erofs_map_blocks *map;
31 	void *kaddr;
32 
33 	unsigned long lcn;
34 	/* compression extent information gathered */
35 	u8  type, headtype;
36 	u16 clusterofs;
37 	u16 delta[2];
38 	erofs_blk_t pblk, compressedblks;
39 	erofs_off_t nextpackoff;
40 	bool partialref;
41 };
42 
43 static int legacy_load_cluster_from_disk(struct z_erofs_maprecorder *m,
44 					 unsigned long lcn)
45 {
46 	struct inode *const inode = m->inode;
47 	struct erofs_inode *const vi = EROFS_I(inode);
48 	const erofs_off_t pos =
49 		Z_EROFS_VLE_LEGACY_INDEX_ALIGN(erofs_iloc(inode) +
50 				vi->inode_isize + vi->xattr_isize) +
51 		lcn * sizeof(struct z_erofs_vle_decompressed_index);
52 	struct z_erofs_vle_decompressed_index *di;
53 	unsigned int advise, type;
54 
55 	m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
56 				      erofs_blknr(pos), EROFS_KMAP);
57 	if (IS_ERR(m->kaddr))
58 		return PTR_ERR(m->kaddr);
59 
60 	m->nextpackoff = pos + sizeof(struct z_erofs_vle_decompressed_index);
61 	m->lcn = lcn;
62 	di = m->kaddr + erofs_blkoff(pos);
63 
64 	advise = le16_to_cpu(di->di_advise);
65 	type = (advise >> Z_EROFS_VLE_DI_CLUSTER_TYPE_BIT) &
66 		((1 << Z_EROFS_VLE_DI_CLUSTER_TYPE_BITS) - 1);
67 	switch (type) {
68 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
69 		m->clusterofs = 1 << vi->z_logical_clusterbits;
70 		m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
71 		if (m->delta[0] & Z_EROFS_VLE_DI_D0_CBLKCNT) {
72 			if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
73 					Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
74 				DBG_BUGON(1);
75 				return -EFSCORRUPTED;
76 			}
77 			m->compressedblks = m->delta[0] &
78 				~Z_EROFS_VLE_DI_D0_CBLKCNT;
79 			m->delta[0] = 1;
80 		}
81 		m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
82 		break;
83 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
84 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
85 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
86 		if (advise & Z_EROFS_VLE_DI_PARTIAL_REF)
87 			m->partialref = true;
88 		m->clusterofs = le16_to_cpu(di->di_clusterofs);
89 		m->pblk = le32_to_cpu(di->di_u.blkaddr);
90 		break;
91 	default:
92 		DBG_BUGON(1);
93 		return -EOPNOTSUPP;
94 	}
95 	m->type = type;
96 	return 0;
97 }
98 
99 static unsigned int decode_compactedbits(unsigned int lobits,
100 					 unsigned int lomask,
101 					 u8 *in, unsigned int pos, u8 *type)
102 {
103 	const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
104 	const unsigned int lo = v & lomask;
105 
106 	*type = (v >> lobits) & 3;
107 	return lo;
108 }
109 
110 static int get_compacted_la_distance(unsigned int lclusterbits,
111 				     unsigned int encodebits,
112 				     unsigned int vcnt, u8 *in, int i)
113 {
114 	const unsigned int lomask = (1 << lclusterbits) - 1;
115 	unsigned int lo, d1 = 0;
116 	u8 type;
117 
118 	DBG_BUGON(i >= vcnt);
119 
120 	do {
121 		lo = decode_compactedbits(lclusterbits, lomask,
122 					  in, encodebits * i, &type);
123 
124 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
125 			return d1;
126 		++d1;
127 	} while (++i < vcnt);
128 
129 	/* vcnt - 1 (Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) item */
130 	if (!(lo & Z_EROFS_VLE_DI_D0_CBLKCNT))
131 		d1 += lo - 1;
132 	return d1;
133 }
134 
135 static int unpack_compacted_index(struct z_erofs_maprecorder *m,
136 				  unsigned int amortizedshift,
137 				  erofs_off_t pos, bool lookahead)
138 {
139 	struct erofs_inode *const vi = EROFS_I(m->inode);
140 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
141 	const unsigned int lomask = (1 << lclusterbits) - 1;
142 	unsigned int vcnt, base, lo, encodebits, nblk, eofs;
143 	int i;
144 	u8 *in, type;
145 	bool big_pcluster;
146 
147 	if (1 << amortizedshift == 4)
148 		vcnt = 2;
149 	else if (1 << amortizedshift == 2 && lclusterbits == 12)
150 		vcnt = 16;
151 	else
152 		return -EOPNOTSUPP;
153 
154 	/* it doesn't equal to round_up(..) */
155 	m->nextpackoff = round_down(pos, vcnt << amortizedshift) +
156 			 (vcnt << amortizedshift);
157 	big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1;
158 	encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
159 	eofs = erofs_blkoff(pos);
160 	base = round_down(eofs, vcnt << amortizedshift);
161 	in = m->kaddr + base;
162 
163 	i = (eofs - base) >> amortizedshift;
164 
165 	lo = decode_compactedbits(lclusterbits, lomask,
166 				  in, encodebits * i, &type);
167 	m->type = type;
168 	if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
169 		m->clusterofs = 1 << lclusterbits;
170 
171 		/* figure out lookahead_distance: delta[1] if needed */
172 		if (lookahead)
173 			m->delta[1] = get_compacted_la_distance(lclusterbits,
174 						encodebits, vcnt, in, i);
175 		if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
176 			if (!big_pcluster) {
177 				DBG_BUGON(1);
178 				return -EFSCORRUPTED;
179 			}
180 			m->compressedblks = lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
181 			m->delta[0] = 1;
182 			return 0;
183 		} else if (i + 1 != (int)vcnt) {
184 			m->delta[0] = lo;
185 			return 0;
186 		}
187 		/*
188 		 * since the last lcluster in the pack is special,
189 		 * of which lo saves delta[1] rather than delta[0].
190 		 * Hence, get delta[0] by the previous lcluster indirectly.
191 		 */
192 		lo = decode_compactedbits(lclusterbits, lomask,
193 					  in, encodebits * (i - 1), &type);
194 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
195 			lo = 0;
196 		else if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT)
197 			lo = 1;
198 		m->delta[0] = lo + 1;
199 		return 0;
200 	}
201 	m->clusterofs = lo;
202 	m->delta[0] = 0;
203 	/* figout out blkaddr (pblk) for HEAD lclusters */
204 	if (!big_pcluster) {
205 		nblk = 1;
206 		while (i > 0) {
207 			--i;
208 			lo = decode_compactedbits(lclusterbits, lomask,
209 						  in, encodebits * i, &type);
210 			if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
211 				i -= lo;
212 
213 			if (i >= 0)
214 				++nblk;
215 		}
216 	} else {
217 		nblk = 0;
218 		while (i > 0) {
219 			--i;
220 			lo = decode_compactedbits(lclusterbits, lomask,
221 						  in, encodebits * i, &type);
222 			if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
223 				if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
224 					--i;
225 					nblk += lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
226 					continue;
227 				}
228 				/* bigpcluster shouldn't have plain d0 == 1 */
229 				if (lo <= 1) {
230 					DBG_BUGON(1);
231 					return -EFSCORRUPTED;
232 				}
233 				i -= lo - 2;
234 				continue;
235 			}
236 			++nblk;
237 		}
238 	}
239 	in += (vcnt << amortizedshift) - sizeof(__le32);
240 	m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
241 	return 0;
242 }
243 
244 static int compacted_load_cluster_from_disk(struct z_erofs_maprecorder *m,
245 					    unsigned long lcn, bool lookahead)
246 {
247 	struct inode *const inode = m->inode;
248 	struct erofs_inode *const vi = EROFS_I(inode);
249 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
250 	const erofs_off_t ebase = sizeof(struct z_erofs_map_header) +
251 		ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8);
252 	const unsigned int totalidx = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
253 	unsigned int compacted_4b_initial, compacted_2b;
254 	unsigned int amortizedshift;
255 	erofs_off_t pos;
256 
257 	if (lclusterbits != 12)
258 		return -EOPNOTSUPP;
259 
260 	if (lcn >= totalidx)
261 		return -EINVAL;
262 
263 	m->lcn = lcn;
264 	/* used to align to 32-byte (compacted_2b) alignment */
265 	compacted_4b_initial = (32 - ebase % 32) / 4;
266 	if (compacted_4b_initial == 32 / 4)
267 		compacted_4b_initial = 0;
268 
269 	if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
270 	    compacted_4b_initial < totalidx)
271 		compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
272 	else
273 		compacted_2b = 0;
274 
275 	pos = ebase;
276 	if (lcn < compacted_4b_initial) {
277 		amortizedshift = 2;
278 		goto out;
279 	}
280 	pos += compacted_4b_initial * 4;
281 	lcn -= compacted_4b_initial;
282 
283 	if (lcn < compacted_2b) {
284 		amortizedshift = 1;
285 		goto out;
286 	}
287 	pos += compacted_2b * 2;
288 	lcn -= compacted_2b;
289 	amortizedshift = 2;
290 out:
291 	pos += lcn * (1 << amortizedshift);
292 	m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
293 				      erofs_blknr(pos), EROFS_KMAP);
294 	if (IS_ERR(m->kaddr))
295 		return PTR_ERR(m->kaddr);
296 	return unpack_compacted_index(m, amortizedshift, pos, lookahead);
297 }
298 
299 static int z_erofs_load_cluster_from_disk(struct z_erofs_maprecorder *m,
300 					  unsigned int lcn, bool lookahead)
301 {
302 	const unsigned int datamode = EROFS_I(m->inode)->datalayout;
303 
304 	if (datamode == EROFS_INODE_FLAT_COMPRESSION_LEGACY)
305 		return legacy_load_cluster_from_disk(m, lcn);
306 
307 	if (datamode == EROFS_INODE_FLAT_COMPRESSION)
308 		return compacted_load_cluster_from_disk(m, lcn, lookahead);
309 
310 	return -EINVAL;
311 }
312 
313 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
314 				   unsigned int lookback_distance)
315 {
316 	struct erofs_inode *const vi = EROFS_I(m->inode);
317 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
318 
319 	while (m->lcn >= lookback_distance) {
320 		unsigned long lcn = m->lcn - lookback_distance;
321 		int err;
322 
323 		/* load extent head logical cluster if needed */
324 		err = z_erofs_load_cluster_from_disk(m, lcn, false);
325 		if (err)
326 			return err;
327 
328 		switch (m->type) {
329 		case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
330 			if (!m->delta[0]) {
331 				erofs_err(m->inode->i_sb,
332 					  "invalid lookback distance 0 @ nid %llu",
333 					  vi->nid);
334 				DBG_BUGON(1);
335 				return -EFSCORRUPTED;
336 			}
337 			lookback_distance = m->delta[0];
338 			continue;
339 		case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
340 		case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
341 		case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
342 			m->headtype = m->type;
343 			m->map->m_la = (lcn << lclusterbits) | m->clusterofs;
344 			return 0;
345 		default:
346 			erofs_err(m->inode->i_sb,
347 				  "unknown type %u @ lcn %lu of nid %llu",
348 				  m->type, lcn, vi->nid);
349 			DBG_BUGON(1);
350 			return -EOPNOTSUPP;
351 		}
352 	}
353 
354 	erofs_err(m->inode->i_sb, "bogus lookback distance @ nid %llu",
355 		  vi->nid);
356 	DBG_BUGON(1);
357 	return -EFSCORRUPTED;
358 }
359 
360 static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m,
361 					    unsigned int initial_lcn)
362 {
363 	struct erofs_inode *const vi = EROFS_I(m->inode);
364 	struct erofs_map_blocks *const map = m->map;
365 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
366 	unsigned long lcn;
367 	int err;
368 
369 	DBG_BUGON(m->type != Z_EROFS_VLE_CLUSTER_TYPE_PLAIN &&
370 		  m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 &&
371 		  m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD2);
372 	DBG_BUGON(m->type != m->headtype);
373 
374 	if (m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
375 	    ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1) &&
376 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1)) ||
377 	    ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) &&
378 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
379 		map->m_plen = 1ULL << lclusterbits;
380 		return 0;
381 	}
382 	lcn = m->lcn + 1;
383 	if (m->compressedblks)
384 		goto out;
385 
386 	err = z_erofs_load_cluster_from_disk(m, lcn, false);
387 	if (err)
388 		return err;
389 
390 	/*
391 	 * If the 1st NONHEAD lcluster has already been handled initially w/o
392 	 * valid compressedblks, which means at least it mustn't be CBLKCNT, or
393 	 * an internal implemenatation error is detected.
394 	 *
395 	 * The following code can also handle it properly anyway, but let's
396 	 * BUG_ON in the debugging mode only for developers to notice that.
397 	 */
398 	DBG_BUGON(lcn == initial_lcn &&
399 		  m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD);
400 
401 	switch (m->type) {
402 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
403 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
404 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
405 		/*
406 		 * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type
407 		 * rather than CBLKCNT, it's a 1 lcluster-sized pcluster.
408 		 */
409 		m->compressedblks = 1 << (lclusterbits - LOG_BLOCK_SIZE);
410 		break;
411 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
412 		if (m->delta[0] != 1)
413 			goto err_bonus_cblkcnt;
414 		if (m->compressedblks)
415 			break;
416 		fallthrough;
417 	default:
418 		erofs_err(m->inode->i_sb,
419 			  "cannot found CBLKCNT @ lcn %lu of nid %llu",
420 			  lcn, vi->nid);
421 		DBG_BUGON(1);
422 		return -EFSCORRUPTED;
423 	}
424 out:
425 	map->m_plen = (u64)m->compressedblks << LOG_BLOCK_SIZE;
426 	return 0;
427 err_bonus_cblkcnt:
428 	erofs_err(m->inode->i_sb,
429 		  "bogus CBLKCNT @ lcn %lu of nid %llu",
430 		  lcn, vi->nid);
431 	DBG_BUGON(1);
432 	return -EFSCORRUPTED;
433 }
434 
435 static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m)
436 {
437 	struct inode *inode = m->inode;
438 	struct erofs_inode *vi = EROFS_I(inode);
439 	struct erofs_map_blocks *map = m->map;
440 	unsigned int lclusterbits = vi->z_logical_clusterbits;
441 	u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits;
442 	int err;
443 
444 	do {
445 		/* handle the last EOF pcluster (no next HEAD lcluster) */
446 		if ((lcn << lclusterbits) >= inode->i_size) {
447 			map->m_llen = inode->i_size - map->m_la;
448 			return 0;
449 		}
450 
451 		err = z_erofs_load_cluster_from_disk(m, lcn, true);
452 		if (err)
453 			return err;
454 
455 		if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
456 			DBG_BUGON(!m->delta[1] &&
457 				  m->clusterofs != 1 << lclusterbits);
458 		} else if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
459 			   m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 ||
460 			   m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) {
461 			/* go on until the next HEAD lcluster */
462 			if (lcn != headlcn)
463 				break;
464 			m->delta[1] = 1;
465 		} else {
466 			erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu",
467 				  m->type, lcn, vi->nid);
468 			DBG_BUGON(1);
469 			return -EOPNOTSUPP;
470 		}
471 		lcn += m->delta[1];
472 	} while (m->delta[1]);
473 
474 	map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la;
475 	return 0;
476 }
477 
478 static int z_erofs_do_map_blocks(struct inode *inode,
479 				 struct erofs_map_blocks *map,
480 				 int flags)
481 {
482 	struct erofs_inode *const vi = EROFS_I(inode);
483 	bool ztailpacking = vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER;
484 	bool fragment = vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
485 	struct z_erofs_maprecorder m = {
486 		.inode = inode,
487 		.map = map,
488 	};
489 	int err = 0;
490 	unsigned int lclusterbits, endoff;
491 	unsigned long initial_lcn;
492 	unsigned long long ofs, end;
493 
494 	lclusterbits = vi->z_logical_clusterbits;
495 	ofs = flags & EROFS_GET_BLOCKS_FINDTAIL ? inode->i_size - 1 : map->m_la;
496 	initial_lcn = ofs >> lclusterbits;
497 	endoff = ofs & ((1 << lclusterbits) - 1);
498 
499 	err = z_erofs_load_cluster_from_disk(&m, initial_lcn, false);
500 	if (err)
501 		goto unmap_out;
502 
503 	if (ztailpacking && (flags & EROFS_GET_BLOCKS_FINDTAIL))
504 		vi->z_idataoff = m.nextpackoff;
505 
506 	map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED;
507 	end = (m.lcn + 1ULL) << lclusterbits;
508 
509 	switch (m.type) {
510 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
511 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
512 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
513 		if (endoff >= m.clusterofs) {
514 			m.headtype = m.type;
515 			map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
516 			/*
517 			 * For ztailpacking files, in order to inline data more
518 			 * effectively, special EOF lclusters are now supported
519 			 * which can have three parts at most.
520 			 */
521 			if (ztailpacking && end > inode->i_size)
522 				end = inode->i_size;
523 			break;
524 		}
525 		/* m.lcn should be >= 1 if endoff < m.clusterofs */
526 		if (!m.lcn) {
527 			erofs_err(inode->i_sb,
528 				  "invalid logical cluster 0 at nid %llu",
529 				  vi->nid);
530 			err = -EFSCORRUPTED;
531 			goto unmap_out;
532 		}
533 		end = (m.lcn << lclusterbits) | m.clusterofs;
534 		map->m_flags |= EROFS_MAP_FULL_MAPPED;
535 		m.delta[0] = 1;
536 		fallthrough;
537 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
538 		/* get the corresponding first chunk */
539 		err = z_erofs_extent_lookback(&m, m.delta[0]);
540 		if (err)
541 			goto unmap_out;
542 		break;
543 	default:
544 		erofs_err(inode->i_sb,
545 			  "unknown type %u @ offset %llu of nid %llu",
546 			  m.type, ofs, vi->nid);
547 		err = -EOPNOTSUPP;
548 		goto unmap_out;
549 	}
550 	if (m.partialref)
551 		map->m_flags |= EROFS_MAP_PARTIAL_REF;
552 	map->m_llen = end - map->m_la;
553 
554 	if (flags & EROFS_GET_BLOCKS_FINDTAIL) {
555 		vi->z_tailextent_headlcn = m.lcn;
556 		/* for non-compact indexes, fragmentoff is 64 bits */
557 		if (fragment &&
558 		    vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY)
559 			vi->z_fragmentoff |= (u64)m.pblk << 32;
560 	}
561 	if (ztailpacking && m.lcn == vi->z_tailextent_headlcn) {
562 		map->m_flags |= EROFS_MAP_META;
563 		map->m_pa = vi->z_idataoff;
564 		map->m_plen = vi->z_idata_size;
565 	} else if (fragment && m.lcn == vi->z_tailextent_headlcn) {
566 		map->m_flags |= EROFS_MAP_FRAGMENT;
567 	} else {
568 		map->m_pa = blknr_to_addr(m.pblk);
569 		err = z_erofs_get_extent_compressedlen(&m, initial_lcn);
570 		if (err)
571 			goto unmap_out;
572 	}
573 
574 	if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN) {
575 		if (map->m_llen > map->m_plen) {
576 			DBG_BUGON(1);
577 			err = -EFSCORRUPTED;
578 			goto unmap_out;
579 		}
580 		if (vi->z_advise & Z_EROFS_ADVISE_INTERLACED_PCLUSTER)
581 			map->m_algorithmformat =
582 				Z_EROFS_COMPRESSION_INTERLACED;
583 		else
584 			map->m_algorithmformat =
585 				Z_EROFS_COMPRESSION_SHIFTED;
586 	} else if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) {
587 		map->m_algorithmformat = vi->z_algorithmtype[1];
588 	} else {
589 		map->m_algorithmformat = vi->z_algorithmtype[0];
590 	}
591 
592 	if ((flags & EROFS_GET_BLOCKS_FIEMAP) ||
593 	    ((flags & EROFS_GET_BLOCKS_READMORE) &&
594 	     map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA &&
595 	     map->m_llen >= EROFS_BLKSIZ)) {
596 		err = z_erofs_get_extent_decompressedlen(&m);
597 		if (!err)
598 			map->m_flags |= EROFS_MAP_FULL_MAPPED;
599 	}
600 
601 unmap_out:
602 	erofs_unmap_metabuf(&m.map->buf);
603 	erofs_dbg("%s, m_la %llu m_pa %llu m_llen %llu m_plen %llu m_flags 0%o",
604 		  __func__, map->m_la, map->m_pa,
605 		  map->m_llen, map->m_plen, map->m_flags);
606 	return err;
607 }
608 
609 static int z_erofs_fill_inode_lazy(struct inode *inode)
610 {
611 	struct erofs_inode *const vi = EROFS_I(inode);
612 	struct super_block *const sb = inode->i_sb;
613 	int err, headnr;
614 	erofs_off_t pos;
615 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
616 	void *kaddr;
617 	struct z_erofs_map_header *h;
618 
619 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) {
620 		/*
621 		 * paired with smp_mb() at the end of the function to ensure
622 		 * fields will only be observed after the bit is set.
623 		 */
624 		smp_mb();
625 		return 0;
626 	}
627 
628 	if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE))
629 		return -ERESTARTSYS;
630 
631 	err = 0;
632 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
633 		goto out_unlock;
634 
635 	pos = ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8);
636 	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP);
637 	if (IS_ERR(kaddr)) {
638 		err = PTR_ERR(kaddr);
639 		goto out_unlock;
640 	}
641 
642 	h = kaddr + erofs_blkoff(pos);
643 	/*
644 	 * if the highest bit of the 8-byte map header is set, the whole file
645 	 * is stored in the packed inode. The rest bits keeps z_fragmentoff.
646 	 */
647 	if (h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT) {
648 		vi->z_advise = Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
649 		vi->z_fragmentoff = le64_to_cpu(*(__le64 *)h) ^ (1ULL << 63);
650 		vi->z_tailextent_headlcn = 0;
651 		goto done;
652 	}
653 	vi->z_advise = le16_to_cpu(h->h_advise);
654 	vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
655 	vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
656 
657 	headnr = 0;
658 	if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX ||
659 	    vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) {
660 		erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel",
661 			  headnr + 1, vi->z_algorithmtype[headnr], vi->nid);
662 		err = -EOPNOTSUPP;
663 		goto out_put_metabuf;
664 	}
665 
666 	vi->z_logical_clusterbits = LOG_BLOCK_SIZE + (h->h_clusterbits & 7);
667 	if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
668 	    vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
669 			    Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
670 		erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu",
671 			  vi->nid);
672 		err = -EFSCORRUPTED;
673 		goto out_put_metabuf;
674 	}
675 	if (vi->datalayout == EROFS_INODE_FLAT_COMPRESSION &&
676 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^
677 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
678 		erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu",
679 			  vi->nid);
680 		err = -EFSCORRUPTED;
681 		goto out_put_metabuf;
682 	}
683 
684 	if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
685 		struct erofs_map_blocks map = {
686 			.buf = __EROFS_BUF_INITIALIZER
687 		};
688 
689 		vi->z_idata_size = le16_to_cpu(h->h_idata_size);
690 		err = z_erofs_do_map_blocks(inode, &map,
691 					    EROFS_GET_BLOCKS_FINDTAIL);
692 		erofs_put_metabuf(&map.buf);
693 
694 		if (!map.m_plen ||
695 		    erofs_blkoff(map.m_pa) + map.m_plen > EROFS_BLKSIZ) {
696 			erofs_err(sb, "invalid tail-packing pclustersize %llu",
697 				  map.m_plen);
698 			err = -EFSCORRUPTED;
699 		}
700 		if (err < 0)
701 			goto out_put_metabuf;
702 	}
703 
704 	if (vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER &&
705 	    !(h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT)) {
706 		struct erofs_map_blocks map = {
707 			.buf = __EROFS_BUF_INITIALIZER
708 		};
709 
710 		vi->z_fragmentoff = le32_to_cpu(h->h_fragmentoff);
711 		err = z_erofs_do_map_blocks(inode, &map,
712 					    EROFS_GET_BLOCKS_FINDTAIL);
713 		erofs_put_metabuf(&map.buf);
714 		if (err < 0)
715 			goto out_put_metabuf;
716 	}
717 done:
718 	/* paired with smp_mb() at the beginning of the function */
719 	smp_mb();
720 	set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
721 out_put_metabuf:
722 	erofs_put_metabuf(&buf);
723 out_unlock:
724 	clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
725 	return err;
726 }
727 
728 int z_erofs_map_blocks_iter(struct inode *inode, struct erofs_map_blocks *map,
729 			    int flags)
730 {
731 	struct erofs_inode *const vi = EROFS_I(inode);
732 	int err = 0;
733 
734 	trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
735 
736 	/* when trying to read beyond EOF, leave it unmapped */
737 	if (map->m_la >= inode->i_size) {
738 		map->m_llen = map->m_la + 1 - inode->i_size;
739 		map->m_la = inode->i_size;
740 		map->m_flags = 0;
741 		goto out;
742 	}
743 
744 	err = z_erofs_fill_inode_lazy(inode);
745 	if (err)
746 		goto out;
747 
748 	if ((vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER) &&
749 	    !vi->z_tailextent_headlcn) {
750 		map->m_la = 0;
751 		map->m_llen = inode->i_size;
752 		map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_FULL_MAPPED |
753 				EROFS_MAP_FRAGMENT;
754 		goto out;
755 	}
756 
757 	err = z_erofs_do_map_blocks(inode, map, flags);
758 out:
759 	trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
760 
761 	/* aggressively BUG_ON iff CONFIG_EROFS_FS_DEBUG is on */
762 	DBG_BUGON(err < 0 && err != -ENOMEM);
763 	return err;
764 }
765 
766 static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
767 				loff_t length, unsigned int flags,
768 				struct iomap *iomap, struct iomap *srcmap)
769 {
770 	int ret;
771 	struct erofs_map_blocks map = { .m_la = offset };
772 
773 	ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
774 	erofs_put_metabuf(&map.buf);
775 	if (ret < 0)
776 		return ret;
777 
778 	iomap->bdev = inode->i_sb->s_bdev;
779 	iomap->offset = map.m_la;
780 	iomap->length = map.m_llen;
781 	if (map.m_flags & EROFS_MAP_MAPPED) {
782 		iomap->type = IOMAP_MAPPED;
783 		iomap->addr = map.m_flags & EROFS_MAP_FRAGMENT ?
784 			      IOMAP_NULL_ADDR : map.m_pa;
785 	} else {
786 		iomap->type = IOMAP_HOLE;
787 		iomap->addr = IOMAP_NULL_ADDR;
788 		/*
789 		 * No strict rule on how to describe extents for post EOF, yet
790 		 * we need to do like below. Otherwise, iomap itself will get
791 		 * into an endless loop on post EOF.
792 		 *
793 		 * Calculate the effective offset by subtracting extent start
794 		 * (map.m_la) from the requested offset, and add it to length.
795 		 * (NB: offset >= map.m_la always)
796 		 */
797 		if (iomap->offset >= inode->i_size)
798 			iomap->length = length + offset - map.m_la;
799 	}
800 	iomap->flags = 0;
801 	return 0;
802 }
803 
804 const struct iomap_ops z_erofs_iomap_report_ops = {
805 	.iomap_begin = z_erofs_iomap_begin_report,
806 };
807