xref: /openbmc/linux/fs/erofs/zmap.c (revision 1fd02f66)
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 static int z_erofs_do_map_blocks(struct inode *inode,
11 				 struct erofs_map_blocks *map,
12 				 int flags);
13 
14 int z_erofs_fill_inode(struct inode *inode)
15 {
16 	struct erofs_inode *const vi = EROFS_I(inode);
17 	struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
18 
19 	if (!erofs_sb_has_big_pcluster(sbi) &&
20 	    !erofs_sb_has_ztailpacking(sbi) &&
21 	    vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY) {
22 		vi->z_advise = 0;
23 		vi->z_algorithmtype[0] = 0;
24 		vi->z_algorithmtype[1] = 0;
25 		vi->z_logical_clusterbits = LOG_BLOCK_SIZE;
26 		set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
27 	}
28 	inode->i_mapping->a_ops = &z_erofs_aops;
29 	return 0;
30 }
31 
32 static int z_erofs_fill_inode_lazy(struct inode *inode)
33 {
34 	struct erofs_inode *const vi = EROFS_I(inode);
35 	struct super_block *const sb = inode->i_sb;
36 	int err, headnr;
37 	erofs_off_t pos;
38 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
39 	void *kaddr;
40 	struct z_erofs_map_header *h;
41 
42 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) {
43 		/*
44 		 * paired with smp_mb() at the end of the function to ensure
45 		 * fields will only be observed after the bit is set.
46 		 */
47 		smp_mb();
48 		return 0;
49 	}
50 
51 	if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE))
52 		return -ERESTARTSYS;
53 
54 	err = 0;
55 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
56 		goto out_unlock;
57 
58 	DBG_BUGON(!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
59 		  !erofs_sb_has_ztailpacking(EROFS_SB(sb)) &&
60 		  vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY);
61 
62 	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
63 		    vi->xattr_isize, 8);
64 	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos),
65 				   EROFS_KMAP_ATOMIC);
66 	if (IS_ERR(kaddr)) {
67 		err = PTR_ERR(kaddr);
68 		goto out_unlock;
69 	}
70 
71 	h = kaddr + erofs_blkoff(pos);
72 	vi->z_advise = le16_to_cpu(h->h_advise);
73 	vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
74 	vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
75 
76 	headnr = 0;
77 	if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX ||
78 	    vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) {
79 		erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel",
80 			  headnr + 1, vi->z_algorithmtype[headnr], vi->nid);
81 		err = -EOPNOTSUPP;
82 		goto unmap_done;
83 	}
84 
85 	vi->z_logical_clusterbits = LOG_BLOCK_SIZE + (h->h_clusterbits & 7);
86 	if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
87 	    vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
88 			    Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
89 		erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu",
90 			  vi->nid);
91 		err = -EFSCORRUPTED;
92 		goto unmap_done;
93 	}
94 	if (vi->datalayout == EROFS_INODE_FLAT_COMPRESSION &&
95 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^
96 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
97 		erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu",
98 			  vi->nid);
99 		err = -EFSCORRUPTED;
100 		goto unmap_done;
101 	}
102 unmap_done:
103 	erofs_put_metabuf(&buf);
104 	if (err)
105 		goto out_unlock;
106 
107 	if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
108 		struct erofs_map_blocks map = {
109 			.buf = __EROFS_BUF_INITIALIZER
110 		};
111 
112 		vi->z_idata_size = le16_to_cpu(h->h_idata_size);
113 		err = z_erofs_do_map_blocks(inode, &map,
114 					    EROFS_GET_BLOCKS_FINDTAIL);
115 		erofs_put_metabuf(&map.buf);
116 
117 		if (!map.m_plen ||
118 		    erofs_blkoff(map.m_pa) + map.m_plen > EROFS_BLKSIZ) {
119 			erofs_err(sb, "invalid tail-packing pclustersize %llu",
120 				  map.m_plen);
121 			err = -EFSCORRUPTED;
122 		}
123 		if (err < 0)
124 			goto out_unlock;
125 	}
126 	/* paired with smp_mb() at the beginning of the function */
127 	smp_mb();
128 	set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
129 out_unlock:
130 	clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
131 	return err;
132 }
133 
134 struct z_erofs_maprecorder {
135 	struct inode *inode;
136 	struct erofs_map_blocks *map;
137 	void *kaddr;
138 
139 	unsigned long lcn;
140 	/* compression extent information gathered */
141 	u8  type, headtype;
142 	u16 clusterofs;
143 	u16 delta[2];
144 	erofs_blk_t pblk, compressedlcs;
145 	erofs_off_t nextpackoff;
146 };
147 
148 static int z_erofs_reload_indexes(struct z_erofs_maprecorder *m,
149 				  erofs_blk_t eblk)
150 {
151 	struct super_block *const sb = m->inode->i_sb;
152 
153 	m->kaddr = erofs_read_metabuf(&m->map->buf, sb, eblk,
154 				      EROFS_KMAP_ATOMIC);
155 	if (IS_ERR(m->kaddr))
156 		return PTR_ERR(m->kaddr);
157 	return 0;
158 }
159 
160 static int legacy_load_cluster_from_disk(struct z_erofs_maprecorder *m,
161 					 unsigned long lcn)
162 {
163 	struct inode *const inode = m->inode;
164 	struct erofs_inode *const vi = EROFS_I(inode);
165 	const erofs_off_t ibase = iloc(EROFS_I_SB(inode), vi->nid);
166 	const erofs_off_t pos =
167 		Z_EROFS_VLE_LEGACY_INDEX_ALIGN(ibase + vi->inode_isize +
168 					       vi->xattr_isize) +
169 		lcn * sizeof(struct z_erofs_vle_decompressed_index);
170 	struct z_erofs_vle_decompressed_index *di;
171 	unsigned int advise, type;
172 	int err;
173 
174 	err = z_erofs_reload_indexes(m, erofs_blknr(pos));
175 	if (err)
176 		return err;
177 
178 	m->nextpackoff = pos + sizeof(struct z_erofs_vle_decompressed_index);
179 	m->lcn = lcn;
180 	di = m->kaddr + erofs_blkoff(pos);
181 
182 	advise = le16_to_cpu(di->di_advise);
183 	type = (advise >> Z_EROFS_VLE_DI_CLUSTER_TYPE_BIT) &
184 		((1 << Z_EROFS_VLE_DI_CLUSTER_TYPE_BITS) - 1);
185 	switch (type) {
186 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
187 		m->clusterofs = 1 << vi->z_logical_clusterbits;
188 		m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
189 		if (m->delta[0] & Z_EROFS_VLE_DI_D0_CBLKCNT) {
190 			if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
191 					Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
192 				DBG_BUGON(1);
193 				return -EFSCORRUPTED;
194 			}
195 			m->compressedlcs = m->delta[0] &
196 				~Z_EROFS_VLE_DI_D0_CBLKCNT;
197 			m->delta[0] = 1;
198 		}
199 		m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
200 		break;
201 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
202 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
203 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
204 		m->clusterofs = le16_to_cpu(di->di_clusterofs);
205 		m->pblk = le32_to_cpu(di->di_u.blkaddr);
206 		break;
207 	default:
208 		DBG_BUGON(1);
209 		return -EOPNOTSUPP;
210 	}
211 	m->type = type;
212 	return 0;
213 }
214 
215 static unsigned int decode_compactedbits(unsigned int lobits,
216 					 unsigned int lomask,
217 					 u8 *in, unsigned int pos, u8 *type)
218 {
219 	const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
220 	const unsigned int lo = v & lomask;
221 
222 	*type = (v >> lobits) & 3;
223 	return lo;
224 }
225 
226 static int get_compacted_la_distance(unsigned int lclusterbits,
227 				     unsigned int encodebits,
228 				     unsigned int vcnt, u8 *in, int i)
229 {
230 	const unsigned int lomask = (1 << lclusterbits) - 1;
231 	unsigned int lo, d1 = 0;
232 	u8 type;
233 
234 	DBG_BUGON(i >= vcnt);
235 
236 	do {
237 		lo = decode_compactedbits(lclusterbits, lomask,
238 					  in, encodebits * i, &type);
239 
240 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
241 			return d1;
242 		++d1;
243 	} while (++i < vcnt);
244 
245 	/* vcnt - 1 (Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) item */
246 	if (!(lo & Z_EROFS_VLE_DI_D0_CBLKCNT))
247 		d1 += lo - 1;
248 	return d1;
249 }
250 
251 static int unpack_compacted_index(struct z_erofs_maprecorder *m,
252 				  unsigned int amortizedshift,
253 				  erofs_off_t pos, bool lookahead)
254 {
255 	struct erofs_inode *const vi = EROFS_I(m->inode);
256 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
257 	const unsigned int lomask = (1 << lclusterbits) - 1;
258 	unsigned int vcnt, base, lo, encodebits, nblk, eofs;
259 	int i;
260 	u8 *in, type;
261 	bool big_pcluster;
262 
263 	if (1 << amortizedshift == 4)
264 		vcnt = 2;
265 	else if (1 << amortizedshift == 2 && lclusterbits == 12)
266 		vcnt = 16;
267 	else
268 		return -EOPNOTSUPP;
269 
270 	/* it doesn't equal to round_up(..) */
271 	m->nextpackoff = round_down(pos, vcnt << amortizedshift) +
272 			 (vcnt << amortizedshift);
273 	big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1;
274 	encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
275 	eofs = erofs_blkoff(pos);
276 	base = round_down(eofs, vcnt << amortizedshift);
277 	in = m->kaddr + base;
278 
279 	i = (eofs - base) >> amortizedshift;
280 
281 	lo = decode_compactedbits(lclusterbits, lomask,
282 				  in, encodebits * i, &type);
283 	m->type = type;
284 	if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
285 		m->clusterofs = 1 << lclusterbits;
286 
287 		/* figure out lookahead_distance: delta[1] if needed */
288 		if (lookahead)
289 			m->delta[1] = get_compacted_la_distance(lclusterbits,
290 						encodebits, vcnt, in, i);
291 		if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
292 			if (!big_pcluster) {
293 				DBG_BUGON(1);
294 				return -EFSCORRUPTED;
295 			}
296 			m->compressedlcs = lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
297 			m->delta[0] = 1;
298 			return 0;
299 		} else if (i + 1 != (int)vcnt) {
300 			m->delta[0] = lo;
301 			return 0;
302 		}
303 		/*
304 		 * since the last lcluster in the pack is special,
305 		 * of which lo saves delta[1] rather than delta[0].
306 		 * Hence, get delta[0] by the previous lcluster indirectly.
307 		 */
308 		lo = decode_compactedbits(lclusterbits, lomask,
309 					  in, encodebits * (i - 1), &type);
310 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
311 			lo = 0;
312 		else if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT)
313 			lo = 1;
314 		m->delta[0] = lo + 1;
315 		return 0;
316 	}
317 	m->clusterofs = lo;
318 	m->delta[0] = 0;
319 	/* figout out blkaddr (pblk) for HEAD lclusters */
320 	if (!big_pcluster) {
321 		nblk = 1;
322 		while (i > 0) {
323 			--i;
324 			lo = decode_compactedbits(lclusterbits, lomask,
325 						  in, encodebits * i, &type);
326 			if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
327 				i -= lo;
328 
329 			if (i >= 0)
330 				++nblk;
331 		}
332 	} else {
333 		nblk = 0;
334 		while (i > 0) {
335 			--i;
336 			lo = decode_compactedbits(lclusterbits, lomask,
337 						  in, encodebits * i, &type);
338 			if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
339 				if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
340 					--i;
341 					nblk += lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
342 					continue;
343 				}
344 				/* bigpcluster shouldn't have plain d0 == 1 */
345 				if (lo <= 1) {
346 					DBG_BUGON(1);
347 					return -EFSCORRUPTED;
348 				}
349 				i -= lo - 2;
350 				continue;
351 			}
352 			++nblk;
353 		}
354 	}
355 	in += (vcnt << amortizedshift) - sizeof(__le32);
356 	m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
357 	return 0;
358 }
359 
360 static int compacted_load_cluster_from_disk(struct z_erofs_maprecorder *m,
361 					    unsigned long lcn, bool lookahead)
362 {
363 	struct inode *const inode = m->inode;
364 	struct erofs_inode *const vi = EROFS_I(inode);
365 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
366 	const erofs_off_t ebase = ALIGN(iloc(EROFS_I_SB(inode), vi->nid) +
367 					vi->inode_isize + vi->xattr_isize, 8) +
368 		sizeof(struct z_erofs_map_header);
369 	const unsigned int totalidx = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
370 	unsigned int compacted_4b_initial, compacted_2b;
371 	unsigned int amortizedshift;
372 	erofs_off_t pos;
373 	int err;
374 
375 	if (lclusterbits != 12)
376 		return -EOPNOTSUPP;
377 
378 	if (lcn >= totalidx)
379 		return -EINVAL;
380 
381 	m->lcn = lcn;
382 	/* used to align to 32-byte (compacted_2b) alignment */
383 	compacted_4b_initial = (32 - ebase % 32) / 4;
384 	if (compacted_4b_initial == 32 / 4)
385 		compacted_4b_initial = 0;
386 
387 	if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
388 	    compacted_4b_initial < totalidx)
389 		compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
390 	else
391 		compacted_2b = 0;
392 
393 	pos = ebase;
394 	if (lcn < compacted_4b_initial) {
395 		amortizedshift = 2;
396 		goto out;
397 	}
398 	pos += compacted_4b_initial * 4;
399 	lcn -= compacted_4b_initial;
400 
401 	if (lcn < compacted_2b) {
402 		amortizedshift = 1;
403 		goto out;
404 	}
405 	pos += compacted_2b * 2;
406 	lcn -= compacted_2b;
407 	amortizedshift = 2;
408 out:
409 	pos += lcn * (1 << amortizedshift);
410 	err = z_erofs_reload_indexes(m, erofs_blknr(pos));
411 	if (err)
412 		return err;
413 	return unpack_compacted_index(m, amortizedshift, pos, lookahead);
414 }
415 
416 static int z_erofs_load_cluster_from_disk(struct z_erofs_maprecorder *m,
417 					  unsigned int lcn, bool lookahead)
418 {
419 	const unsigned int datamode = EROFS_I(m->inode)->datalayout;
420 
421 	if (datamode == EROFS_INODE_FLAT_COMPRESSION_LEGACY)
422 		return legacy_load_cluster_from_disk(m, lcn);
423 
424 	if (datamode == EROFS_INODE_FLAT_COMPRESSION)
425 		return compacted_load_cluster_from_disk(m, lcn, lookahead);
426 
427 	return -EINVAL;
428 }
429 
430 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
431 				   unsigned int lookback_distance)
432 {
433 	struct erofs_inode *const vi = EROFS_I(m->inode);
434 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
435 
436 	while (m->lcn >= lookback_distance) {
437 		unsigned long lcn = m->lcn - lookback_distance;
438 		int err;
439 
440 		/* load extent head logical cluster if needed */
441 		err = z_erofs_load_cluster_from_disk(m, lcn, false);
442 		if (err)
443 			return err;
444 
445 		switch (m->type) {
446 		case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
447 			if (!m->delta[0]) {
448 				erofs_err(m->inode->i_sb,
449 					  "invalid lookback distance 0 @ nid %llu",
450 					  vi->nid);
451 				DBG_BUGON(1);
452 				return -EFSCORRUPTED;
453 			}
454 			lookback_distance = m->delta[0];
455 			continue;
456 		case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
457 		case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
458 		case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
459 			m->headtype = m->type;
460 			m->map->m_la = (lcn << lclusterbits) | m->clusterofs;
461 			return 0;
462 		default:
463 			erofs_err(m->inode->i_sb,
464 				  "unknown type %u @ lcn %lu of nid %llu",
465 				  m->type, lcn, vi->nid);
466 			DBG_BUGON(1);
467 			return -EOPNOTSUPP;
468 		}
469 	}
470 
471 	erofs_err(m->inode->i_sb, "bogus lookback distance @ nid %llu",
472 		  vi->nid);
473 	DBG_BUGON(1);
474 	return -EFSCORRUPTED;
475 }
476 
477 static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m,
478 					    unsigned int initial_lcn)
479 {
480 	struct erofs_inode *const vi = EROFS_I(m->inode);
481 	struct erofs_map_blocks *const map = m->map;
482 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
483 	unsigned long lcn;
484 	int err;
485 
486 	DBG_BUGON(m->type != Z_EROFS_VLE_CLUSTER_TYPE_PLAIN &&
487 		  m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 &&
488 		  m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD2);
489 	DBG_BUGON(m->type != m->headtype);
490 
491 	if (m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
492 	    ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1) &&
493 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1)) ||
494 	    ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) &&
495 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
496 		map->m_plen = 1ULL << lclusterbits;
497 		return 0;
498 	}
499 	lcn = m->lcn + 1;
500 	if (m->compressedlcs)
501 		goto out;
502 
503 	err = z_erofs_load_cluster_from_disk(m, lcn, false);
504 	if (err)
505 		return err;
506 
507 	/*
508 	 * If the 1st NONHEAD lcluster has already been handled initially w/o
509 	 * valid compressedlcs, which means at least it mustn't be CBLKCNT, or
510 	 * an internal implemenatation error is detected.
511 	 *
512 	 * The following code can also handle it properly anyway, but let's
513 	 * BUG_ON in the debugging mode only for developers to notice that.
514 	 */
515 	DBG_BUGON(lcn == initial_lcn &&
516 		  m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD);
517 
518 	switch (m->type) {
519 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
520 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
521 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
522 		/*
523 		 * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type
524 		 * rather than CBLKCNT, it's a 1 lcluster-sized pcluster.
525 		 */
526 		m->compressedlcs = 1;
527 		break;
528 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
529 		if (m->delta[0] != 1)
530 			goto err_bonus_cblkcnt;
531 		if (m->compressedlcs)
532 			break;
533 		fallthrough;
534 	default:
535 		erofs_err(m->inode->i_sb,
536 			  "cannot found CBLKCNT @ lcn %lu of nid %llu",
537 			  lcn, vi->nid);
538 		DBG_BUGON(1);
539 		return -EFSCORRUPTED;
540 	}
541 out:
542 	map->m_plen = (u64)m->compressedlcs << lclusterbits;
543 	return 0;
544 err_bonus_cblkcnt:
545 	erofs_err(m->inode->i_sb,
546 		  "bogus CBLKCNT @ lcn %lu of nid %llu",
547 		  lcn, vi->nid);
548 	DBG_BUGON(1);
549 	return -EFSCORRUPTED;
550 }
551 
552 static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m)
553 {
554 	struct inode *inode = m->inode;
555 	struct erofs_inode *vi = EROFS_I(inode);
556 	struct erofs_map_blocks *map = m->map;
557 	unsigned int lclusterbits = vi->z_logical_clusterbits;
558 	u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits;
559 	int err;
560 
561 	do {
562 		/* handle the last EOF pcluster (no next HEAD lcluster) */
563 		if ((lcn << lclusterbits) >= inode->i_size) {
564 			map->m_llen = inode->i_size - map->m_la;
565 			return 0;
566 		}
567 
568 		err = z_erofs_load_cluster_from_disk(m, lcn, true);
569 		if (err)
570 			return err;
571 
572 		if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
573 			DBG_BUGON(!m->delta[1] &&
574 				  m->clusterofs != 1 << lclusterbits);
575 		} else if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
576 			   m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 ||
577 			   m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) {
578 			/* go on until the next HEAD lcluster */
579 			if (lcn != headlcn)
580 				break;
581 			m->delta[1] = 1;
582 		} else {
583 			erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu",
584 				  m->type, lcn, vi->nid);
585 			DBG_BUGON(1);
586 			return -EOPNOTSUPP;
587 		}
588 		lcn += m->delta[1];
589 	} while (m->delta[1]);
590 
591 	map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la;
592 	return 0;
593 }
594 
595 static int z_erofs_do_map_blocks(struct inode *inode,
596 				 struct erofs_map_blocks *map,
597 				 int flags)
598 {
599 	struct erofs_inode *const vi = EROFS_I(inode);
600 	bool ztailpacking = vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER;
601 	struct z_erofs_maprecorder m = {
602 		.inode = inode,
603 		.map = map,
604 	};
605 	int err = 0;
606 	unsigned int lclusterbits, endoff;
607 	unsigned long initial_lcn;
608 	unsigned long long ofs, end;
609 
610 	lclusterbits = vi->z_logical_clusterbits;
611 	ofs = flags & EROFS_GET_BLOCKS_FINDTAIL ? inode->i_size - 1 : map->m_la;
612 	initial_lcn = ofs >> lclusterbits;
613 	endoff = ofs & ((1 << lclusterbits) - 1);
614 
615 	err = z_erofs_load_cluster_from_disk(&m, initial_lcn, false);
616 	if (err)
617 		goto unmap_out;
618 
619 	if (ztailpacking && (flags & EROFS_GET_BLOCKS_FINDTAIL))
620 		vi->z_idataoff = m.nextpackoff;
621 
622 	map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED;
623 	end = (m.lcn + 1ULL) << lclusterbits;
624 
625 	switch (m.type) {
626 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
627 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
628 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
629 		if (endoff >= m.clusterofs) {
630 			m.headtype = m.type;
631 			map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
632 			/*
633 			 * For ztailpacking files, in order to inline data more
634 			 * effectively, special EOF lclusters are now supported
635 			 * which can have three parts at most.
636 			 */
637 			if (ztailpacking && end > inode->i_size)
638 				end = inode->i_size;
639 			break;
640 		}
641 		/* m.lcn should be >= 1 if endoff < m.clusterofs */
642 		if (!m.lcn) {
643 			erofs_err(inode->i_sb,
644 				  "invalid logical cluster 0 at nid %llu",
645 				  vi->nid);
646 			err = -EFSCORRUPTED;
647 			goto unmap_out;
648 		}
649 		end = (m.lcn << lclusterbits) | m.clusterofs;
650 		map->m_flags |= EROFS_MAP_FULL_MAPPED;
651 		m.delta[0] = 1;
652 		fallthrough;
653 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
654 		/* get the corresponding first chunk */
655 		err = z_erofs_extent_lookback(&m, m.delta[0]);
656 		if (err)
657 			goto unmap_out;
658 		break;
659 	default:
660 		erofs_err(inode->i_sb,
661 			  "unknown type %u @ offset %llu of nid %llu",
662 			  m.type, ofs, vi->nid);
663 		err = -EOPNOTSUPP;
664 		goto unmap_out;
665 	}
666 
667 	map->m_llen = end - map->m_la;
668 
669 	if (flags & EROFS_GET_BLOCKS_FINDTAIL)
670 		vi->z_tailextent_headlcn = m.lcn;
671 	if (ztailpacking && m.lcn == vi->z_tailextent_headlcn) {
672 		map->m_flags |= EROFS_MAP_META;
673 		map->m_pa = vi->z_idataoff;
674 		map->m_plen = vi->z_idata_size;
675 	} else {
676 		map->m_pa = blknr_to_addr(m.pblk);
677 		err = z_erofs_get_extent_compressedlen(&m, initial_lcn);
678 		if (err)
679 			goto out;
680 	}
681 
682 	if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN)
683 		map->m_algorithmformat = Z_EROFS_COMPRESSION_SHIFTED;
684 	else if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2)
685 		map->m_algorithmformat = vi->z_algorithmtype[1];
686 	else
687 		map->m_algorithmformat = vi->z_algorithmtype[0];
688 
689 	if ((flags & EROFS_GET_BLOCKS_FIEMAP) ||
690 	    ((flags & EROFS_GET_BLOCKS_READMORE) &&
691 	     map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA &&
692 	     map->m_llen >= EROFS_BLKSIZ)) {
693 		err = z_erofs_get_extent_decompressedlen(&m);
694 		if (!err)
695 			map->m_flags |= EROFS_MAP_FULL_MAPPED;
696 	}
697 unmap_out:
698 	erofs_unmap_metabuf(&m.map->buf);
699 
700 out:
701 	erofs_dbg("%s, m_la %llu m_pa %llu m_llen %llu m_plen %llu m_flags 0%o",
702 		  __func__, map->m_la, map->m_pa,
703 		  map->m_llen, map->m_plen, map->m_flags);
704 
705 	return err;
706 }
707 
708 int z_erofs_map_blocks_iter(struct inode *inode,
709 			    struct erofs_map_blocks *map,
710 			    int flags)
711 {
712 	int err = 0;
713 
714 	trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
715 
716 	/* when trying to read beyond EOF, leave it unmapped */
717 	if (map->m_la >= inode->i_size) {
718 		map->m_llen = map->m_la + 1 - inode->i_size;
719 		map->m_la = inode->i_size;
720 		map->m_flags = 0;
721 		goto out;
722 	}
723 
724 	err = z_erofs_fill_inode_lazy(inode);
725 	if (err)
726 		goto out;
727 
728 	err = z_erofs_do_map_blocks(inode, map, flags);
729 out:
730 	trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
731 
732 	/* aggressively BUG_ON iff CONFIG_EROFS_FS_DEBUG is on */
733 	DBG_BUGON(err < 0 && err != -ENOMEM);
734 	return err;
735 }
736 
737 static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
738 				loff_t length, unsigned int flags,
739 				struct iomap *iomap, struct iomap *srcmap)
740 {
741 	int ret;
742 	struct erofs_map_blocks map = { .m_la = offset };
743 
744 	ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
745 	erofs_put_metabuf(&map.buf);
746 	if (ret < 0)
747 		return ret;
748 
749 	iomap->bdev = inode->i_sb->s_bdev;
750 	iomap->offset = map.m_la;
751 	iomap->length = map.m_llen;
752 	if (map.m_flags & EROFS_MAP_MAPPED) {
753 		iomap->type = IOMAP_MAPPED;
754 		iomap->addr = map.m_pa;
755 	} else {
756 		iomap->type = IOMAP_HOLE;
757 		iomap->addr = IOMAP_NULL_ADDR;
758 		/*
759 		 * No strict rule how to describe extents for post EOF, yet
760 		 * we need do like below. Otherwise, iomap itself will get
761 		 * into an endless loop on post EOF.
762 		 */
763 		if (iomap->offset >= inode->i_size)
764 			iomap->length = length + map.m_la - offset;
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