xref: /openbmc/linux/fs/erofs/zmap.c (revision 6d425d7c)
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 	    vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY) {
17 		vi->z_advise = 0;
18 		vi->z_algorithmtype[0] = 0;
19 		vi->z_algorithmtype[1] = 0;
20 		vi->z_logical_clusterbits = LOG_BLOCK_SIZE;
21 		set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
22 	}
23 	inode->i_mapping->a_ops = &z_erofs_aops;
24 	return 0;
25 }
26 
27 static int z_erofs_fill_inode_lazy(struct inode *inode)
28 {
29 	struct erofs_inode *const vi = EROFS_I(inode);
30 	struct super_block *const sb = inode->i_sb;
31 	int err, headnr;
32 	erofs_off_t pos;
33 	struct page *page;
34 	void *kaddr;
35 	struct z_erofs_map_header *h;
36 
37 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) {
38 		/*
39 		 * paired with smp_mb() at the end of the function to ensure
40 		 * fields will only be observed after the bit is set.
41 		 */
42 		smp_mb();
43 		return 0;
44 	}
45 
46 	if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE))
47 		return -ERESTARTSYS;
48 
49 	err = 0;
50 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
51 		goto out_unlock;
52 
53 	DBG_BUGON(!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
54 		  vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY);
55 
56 	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
57 		    vi->xattr_isize, 8);
58 	page = erofs_get_meta_page(sb, erofs_blknr(pos));
59 	if (IS_ERR(page)) {
60 		err = PTR_ERR(page);
61 		goto out_unlock;
62 	}
63 
64 	kaddr = kmap_atomic(page);
65 
66 	h = kaddr + erofs_blkoff(pos);
67 	vi->z_advise = le16_to_cpu(h->h_advise);
68 	vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
69 	vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
70 
71 	headnr = 0;
72 	if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX ||
73 	    vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) {
74 		erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel",
75 			  headnr + 1, vi->z_algorithmtype[headnr], vi->nid);
76 		err = -EOPNOTSUPP;
77 		goto unmap_done;
78 	}
79 
80 	vi->z_logical_clusterbits = LOG_BLOCK_SIZE + (h->h_clusterbits & 7);
81 	if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
82 	    vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
83 			    Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
84 		erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu",
85 			  vi->nid);
86 		err = -EFSCORRUPTED;
87 		goto unmap_done;
88 	}
89 	if (vi->datalayout == EROFS_INODE_FLAT_COMPRESSION &&
90 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^
91 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
92 		erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu",
93 			  vi->nid);
94 		err = -EFSCORRUPTED;
95 		goto unmap_done;
96 	}
97 	/* paired with smp_mb() at the beginning of the function */
98 	smp_mb();
99 	set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
100 unmap_done:
101 	kunmap_atomic(kaddr);
102 	unlock_page(page);
103 	put_page(page);
104 out_unlock:
105 	clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
106 	return err;
107 }
108 
109 struct z_erofs_maprecorder {
110 	struct inode *inode;
111 	struct erofs_map_blocks *map;
112 	void *kaddr;
113 
114 	unsigned long lcn;
115 	/* compression extent information gathered */
116 	u8  type, headtype;
117 	u16 clusterofs;
118 	u16 delta[2];
119 	erofs_blk_t pblk, compressedlcs;
120 };
121 
122 static int z_erofs_reload_indexes(struct z_erofs_maprecorder *m,
123 				  erofs_blk_t eblk)
124 {
125 	struct super_block *const sb = m->inode->i_sb;
126 	struct erofs_map_blocks *const map = m->map;
127 	struct page *mpage = map->mpage;
128 
129 	if (mpage) {
130 		if (mpage->index == eblk) {
131 			if (!m->kaddr)
132 				m->kaddr = kmap_atomic(mpage);
133 			return 0;
134 		}
135 
136 		if (m->kaddr) {
137 			kunmap_atomic(m->kaddr);
138 			m->kaddr = NULL;
139 		}
140 		put_page(mpage);
141 	}
142 
143 	mpage = erofs_get_meta_page(sb, eblk);
144 	if (IS_ERR(mpage)) {
145 		map->mpage = NULL;
146 		return PTR_ERR(mpage);
147 	}
148 	m->kaddr = kmap_atomic(mpage);
149 	unlock_page(mpage);
150 	map->mpage = mpage;
151 	return 0;
152 }
153 
154 static int legacy_load_cluster_from_disk(struct z_erofs_maprecorder *m,
155 					 unsigned long lcn)
156 {
157 	struct inode *const inode = m->inode;
158 	struct erofs_inode *const vi = EROFS_I(inode);
159 	const erofs_off_t ibase = iloc(EROFS_I_SB(inode), vi->nid);
160 	const erofs_off_t pos =
161 		Z_EROFS_VLE_LEGACY_INDEX_ALIGN(ibase + vi->inode_isize +
162 					       vi->xattr_isize) +
163 		lcn * sizeof(struct z_erofs_vle_decompressed_index);
164 	struct z_erofs_vle_decompressed_index *di;
165 	unsigned int advise, type;
166 	int err;
167 
168 	err = z_erofs_reload_indexes(m, erofs_blknr(pos));
169 	if (err)
170 		return err;
171 
172 	m->lcn = lcn;
173 	di = m->kaddr + erofs_blkoff(pos);
174 
175 	advise = le16_to_cpu(di->di_advise);
176 	type = (advise >> Z_EROFS_VLE_DI_CLUSTER_TYPE_BIT) &
177 		((1 << Z_EROFS_VLE_DI_CLUSTER_TYPE_BITS) - 1);
178 	switch (type) {
179 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
180 		m->clusterofs = 1 << vi->z_logical_clusterbits;
181 		m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
182 		if (m->delta[0] & Z_EROFS_VLE_DI_D0_CBLKCNT) {
183 			if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
184 					Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
185 				DBG_BUGON(1);
186 				return -EFSCORRUPTED;
187 			}
188 			m->compressedlcs = m->delta[0] &
189 				~Z_EROFS_VLE_DI_D0_CBLKCNT;
190 			m->delta[0] = 1;
191 		}
192 		m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
193 		break;
194 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
195 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
196 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
197 		m->clusterofs = le16_to_cpu(di->di_clusterofs);
198 		m->pblk = le32_to_cpu(di->di_u.blkaddr);
199 		break;
200 	default:
201 		DBG_BUGON(1);
202 		return -EOPNOTSUPP;
203 	}
204 	m->type = type;
205 	return 0;
206 }
207 
208 static unsigned int decode_compactedbits(unsigned int lobits,
209 					 unsigned int lomask,
210 					 u8 *in, unsigned int pos, u8 *type)
211 {
212 	const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
213 	const unsigned int lo = v & lomask;
214 
215 	*type = (v >> lobits) & 3;
216 	return lo;
217 }
218 
219 static int get_compacted_la_distance(unsigned int lclusterbits,
220 				     unsigned int encodebits,
221 				     unsigned int vcnt, u8 *in, int i)
222 {
223 	const unsigned int lomask = (1 << lclusterbits) - 1;
224 	unsigned int lo, d1 = 0;
225 	u8 type;
226 
227 	DBG_BUGON(i >= vcnt);
228 
229 	do {
230 		lo = decode_compactedbits(lclusterbits, lomask,
231 					  in, encodebits * i, &type);
232 
233 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
234 			return d1;
235 		++d1;
236 	} while (++i < vcnt);
237 
238 	/* vcnt - 1 (Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) item */
239 	if (!(lo & Z_EROFS_VLE_DI_D0_CBLKCNT))
240 		d1 += lo - 1;
241 	return d1;
242 }
243 
244 static int unpack_compacted_index(struct z_erofs_maprecorder *m,
245 				  unsigned int amortizedshift,
246 				  unsigned int eofs, bool lookahead)
247 {
248 	struct erofs_inode *const vi = EROFS_I(m->inode);
249 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
250 	const unsigned int lomask = (1 << lclusterbits) - 1;
251 	unsigned int vcnt, base, lo, encodebits, nblk;
252 	int i;
253 	u8 *in, type;
254 	bool big_pcluster;
255 
256 	if (1 << amortizedshift == 4)
257 		vcnt = 2;
258 	else if (1 << amortizedshift == 2 && lclusterbits == 12)
259 		vcnt = 16;
260 	else
261 		return -EOPNOTSUPP;
262 
263 	big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1;
264 	encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
265 	base = round_down(eofs, vcnt << amortizedshift);
266 	in = m->kaddr + base;
267 
268 	i = (eofs - base) >> amortizedshift;
269 
270 	lo = decode_compactedbits(lclusterbits, lomask,
271 				  in, encodebits * i, &type);
272 	m->type = type;
273 	if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
274 		m->clusterofs = 1 << lclusterbits;
275 
276 		/* figure out lookahead_distance: delta[1] if needed */
277 		if (lookahead)
278 			m->delta[1] = get_compacted_la_distance(lclusterbits,
279 						encodebits, vcnt, in, i);
280 		if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
281 			if (!big_pcluster) {
282 				DBG_BUGON(1);
283 				return -EFSCORRUPTED;
284 			}
285 			m->compressedlcs = lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
286 			m->delta[0] = 1;
287 			return 0;
288 		} else if (i + 1 != (int)vcnt) {
289 			m->delta[0] = lo;
290 			return 0;
291 		}
292 		/*
293 		 * since the last lcluster in the pack is special,
294 		 * of which lo saves delta[1] rather than delta[0].
295 		 * Hence, get delta[0] by the previous lcluster indirectly.
296 		 */
297 		lo = decode_compactedbits(lclusterbits, lomask,
298 					  in, encodebits * (i - 1), &type);
299 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
300 			lo = 0;
301 		else if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT)
302 			lo = 1;
303 		m->delta[0] = lo + 1;
304 		return 0;
305 	}
306 	m->clusterofs = lo;
307 	m->delta[0] = 0;
308 	/* figout out blkaddr (pblk) for HEAD lclusters */
309 	if (!big_pcluster) {
310 		nblk = 1;
311 		while (i > 0) {
312 			--i;
313 			lo = decode_compactedbits(lclusterbits, lomask,
314 						  in, encodebits * i, &type);
315 			if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
316 				i -= lo;
317 
318 			if (i >= 0)
319 				++nblk;
320 		}
321 	} else {
322 		nblk = 0;
323 		while (i > 0) {
324 			--i;
325 			lo = decode_compactedbits(lclusterbits, lomask,
326 						  in, encodebits * i, &type);
327 			if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
328 				if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
329 					--i;
330 					nblk += lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
331 					continue;
332 				}
333 				/* bigpcluster shouldn't have plain d0 == 1 */
334 				if (lo <= 1) {
335 					DBG_BUGON(1);
336 					return -EFSCORRUPTED;
337 				}
338 				i -= lo - 2;
339 				continue;
340 			}
341 			++nblk;
342 		}
343 	}
344 	in += (vcnt << amortizedshift) - sizeof(__le32);
345 	m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
346 	return 0;
347 }
348 
349 static int compacted_load_cluster_from_disk(struct z_erofs_maprecorder *m,
350 					    unsigned long lcn, bool lookahead)
351 {
352 	struct inode *const inode = m->inode;
353 	struct erofs_inode *const vi = EROFS_I(inode);
354 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
355 	const erofs_off_t ebase = ALIGN(iloc(EROFS_I_SB(inode), vi->nid) +
356 					vi->inode_isize + vi->xattr_isize, 8) +
357 		sizeof(struct z_erofs_map_header);
358 	const unsigned int totalidx = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
359 	unsigned int compacted_4b_initial, compacted_2b;
360 	unsigned int amortizedshift;
361 	erofs_off_t pos;
362 	int err;
363 
364 	if (lclusterbits != 12)
365 		return -EOPNOTSUPP;
366 
367 	if (lcn >= totalidx)
368 		return -EINVAL;
369 
370 	m->lcn = lcn;
371 	/* used to align to 32-byte (compacted_2b) alignment */
372 	compacted_4b_initial = (32 - ebase % 32) / 4;
373 	if (compacted_4b_initial == 32 / 4)
374 		compacted_4b_initial = 0;
375 
376 	if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
377 	    compacted_4b_initial < totalidx)
378 		compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
379 	else
380 		compacted_2b = 0;
381 
382 	pos = ebase;
383 	if (lcn < compacted_4b_initial) {
384 		amortizedshift = 2;
385 		goto out;
386 	}
387 	pos += compacted_4b_initial * 4;
388 	lcn -= compacted_4b_initial;
389 
390 	if (lcn < compacted_2b) {
391 		amortizedshift = 1;
392 		goto out;
393 	}
394 	pos += compacted_2b * 2;
395 	lcn -= compacted_2b;
396 	amortizedshift = 2;
397 out:
398 	pos += lcn * (1 << amortizedshift);
399 	err = z_erofs_reload_indexes(m, erofs_blknr(pos));
400 	if (err)
401 		return err;
402 	return unpack_compacted_index(m, amortizedshift, erofs_blkoff(pos),
403 				      lookahead);
404 }
405 
406 static int z_erofs_load_cluster_from_disk(struct z_erofs_maprecorder *m,
407 					  unsigned int lcn, bool lookahead)
408 {
409 	const unsigned int datamode = EROFS_I(m->inode)->datalayout;
410 
411 	if (datamode == EROFS_INODE_FLAT_COMPRESSION_LEGACY)
412 		return legacy_load_cluster_from_disk(m, lcn);
413 
414 	if (datamode == EROFS_INODE_FLAT_COMPRESSION)
415 		return compacted_load_cluster_from_disk(m, lcn, lookahead);
416 
417 	return -EINVAL;
418 }
419 
420 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
421 				   unsigned int lookback_distance)
422 {
423 	struct erofs_inode *const vi = EROFS_I(m->inode);
424 	struct erofs_map_blocks *const map = m->map;
425 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
426 	unsigned long lcn = m->lcn;
427 	int err;
428 
429 	if (lcn < lookback_distance) {
430 		erofs_err(m->inode->i_sb,
431 			  "bogus lookback distance @ nid %llu", vi->nid);
432 		DBG_BUGON(1);
433 		return -EFSCORRUPTED;
434 	}
435 
436 	/* load extent head logical cluster if needed */
437 	lcn -= lookback_distance;
438 	err = z_erofs_load_cluster_from_disk(m, lcn, false);
439 	if (err)
440 		return err;
441 
442 	switch (m->type) {
443 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
444 		if (!m->delta[0]) {
445 			erofs_err(m->inode->i_sb,
446 				  "invalid lookback distance 0 @ nid %llu",
447 				  vi->nid);
448 			DBG_BUGON(1);
449 			return -EFSCORRUPTED;
450 		}
451 		return z_erofs_extent_lookback(m, m->delta[0]);
452 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
453 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
454 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
455 		m->headtype = m->type;
456 		map->m_la = (lcn << lclusterbits) | m->clusterofs;
457 		break;
458 	default:
459 		erofs_err(m->inode->i_sb,
460 			  "unknown type %u @ lcn %lu of nid %llu",
461 			  m->type, lcn, vi->nid);
462 		DBG_BUGON(1);
463 		return -EOPNOTSUPP;
464 	}
465 	return 0;
466 }
467 
468 static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m,
469 					    unsigned int initial_lcn)
470 {
471 	struct erofs_inode *const vi = EROFS_I(m->inode);
472 	struct erofs_map_blocks *const map = m->map;
473 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
474 	unsigned long lcn;
475 	int err;
476 
477 	DBG_BUGON(m->type != Z_EROFS_VLE_CLUSTER_TYPE_PLAIN &&
478 		  m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 &&
479 		  m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD2);
480 	DBG_BUGON(m->type != m->headtype);
481 
482 	if (m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
483 	    ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1) &&
484 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1)) ||
485 	    ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) &&
486 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
487 		map->m_plen = 1 << lclusterbits;
488 		return 0;
489 	}
490 	lcn = m->lcn + 1;
491 	if (m->compressedlcs)
492 		goto out;
493 
494 	err = z_erofs_load_cluster_from_disk(m, lcn, false);
495 	if (err)
496 		return err;
497 
498 	/*
499 	 * If the 1st NONHEAD lcluster has already been handled initially w/o
500 	 * valid compressedlcs, which means at least it mustn't be CBLKCNT, or
501 	 * an internal implemenatation error is detected.
502 	 *
503 	 * The following code can also handle it properly anyway, but let's
504 	 * BUG_ON in the debugging mode only for developers to notice that.
505 	 */
506 	DBG_BUGON(lcn == initial_lcn &&
507 		  m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD);
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 		/*
514 		 * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type
515 		 * rather than CBLKCNT, it's a 1 lcluster-sized pcluster.
516 		 */
517 		m->compressedlcs = 1;
518 		break;
519 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
520 		if (m->delta[0] != 1)
521 			goto err_bonus_cblkcnt;
522 		if (m->compressedlcs)
523 			break;
524 		fallthrough;
525 	default:
526 		erofs_err(m->inode->i_sb,
527 			  "cannot found CBLKCNT @ lcn %lu of nid %llu",
528 			  lcn, vi->nid);
529 		DBG_BUGON(1);
530 		return -EFSCORRUPTED;
531 	}
532 out:
533 	map->m_plen = m->compressedlcs << lclusterbits;
534 	return 0;
535 err_bonus_cblkcnt:
536 	erofs_err(m->inode->i_sb,
537 		  "bogus CBLKCNT @ lcn %lu of nid %llu",
538 		  lcn, vi->nid);
539 	DBG_BUGON(1);
540 	return -EFSCORRUPTED;
541 }
542 
543 static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m)
544 {
545 	struct inode *inode = m->inode;
546 	struct erofs_inode *vi = EROFS_I(inode);
547 	struct erofs_map_blocks *map = m->map;
548 	unsigned int lclusterbits = vi->z_logical_clusterbits;
549 	u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits;
550 	int err;
551 
552 	do {
553 		/* handle the last EOF pcluster (no next HEAD lcluster) */
554 		if ((lcn << lclusterbits) >= inode->i_size) {
555 			map->m_llen = inode->i_size - map->m_la;
556 			return 0;
557 		}
558 
559 		err = z_erofs_load_cluster_from_disk(m, lcn, true);
560 		if (err)
561 			return err;
562 
563 		if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
564 			DBG_BUGON(!m->delta[1] &&
565 				  m->clusterofs != 1 << lclusterbits);
566 		} else if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
567 			   m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 ||
568 			   m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) {
569 			/* go on until the next HEAD lcluster */
570 			if (lcn != headlcn)
571 				break;
572 			m->delta[1] = 1;
573 		} else {
574 			erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu",
575 				  m->type, lcn, vi->nid);
576 			DBG_BUGON(1);
577 			return -EOPNOTSUPP;
578 		}
579 		lcn += m->delta[1];
580 	} while (m->delta[1]);
581 
582 	map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la;
583 	return 0;
584 }
585 
586 int z_erofs_map_blocks_iter(struct inode *inode,
587 			    struct erofs_map_blocks *map,
588 			    int flags)
589 {
590 	struct erofs_inode *const vi = EROFS_I(inode);
591 	struct z_erofs_maprecorder m = {
592 		.inode = inode,
593 		.map = map,
594 	};
595 	int err = 0;
596 	unsigned int lclusterbits, endoff;
597 	unsigned long initial_lcn;
598 	unsigned long long ofs, end;
599 
600 	trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
601 
602 	/* when trying to read beyond EOF, leave it unmapped */
603 	if (map->m_la >= inode->i_size) {
604 		map->m_llen = map->m_la + 1 - inode->i_size;
605 		map->m_la = inode->i_size;
606 		map->m_flags = 0;
607 		goto out;
608 	}
609 
610 	err = z_erofs_fill_inode_lazy(inode);
611 	if (err)
612 		goto out;
613 
614 	lclusterbits = vi->z_logical_clusterbits;
615 	ofs = map->m_la;
616 	initial_lcn = ofs >> lclusterbits;
617 	endoff = ofs & ((1 << lclusterbits) - 1);
618 
619 	err = z_erofs_load_cluster_from_disk(&m, initial_lcn, false);
620 	if (err)
621 		goto unmap_out;
622 
623 	map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED;
624 	end = (m.lcn + 1ULL) << lclusterbits;
625 
626 	switch (m.type) {
627 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
628 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
629 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
630 		if (endoff >= m.clusterofs) {
631 			m.headtype = m.type;
632 			map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
633 			break;
634 		}
635 		/* m.lcn should be >= 1 if endoff < m.clusterofs */
636 		if (!m.lcn) {
637 			erofs_err(inode->i_sb,
638 				  "invalid logical cluster 0 at nid %llu",
639 				  vi->nid);
640 			err = -EFSCORRUPTED;
641 			goto unmap_out;
642 		}
643 		end = (m.lcn << lclusterbits) | m.clusterofs;
644 		map->m_flags |= EROFS_MAP_FULL_MAPPED;
645 		m.delta[0] = 1;
646 		fallthrough;
647 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
648 		/* get the corresponding first chunk */
649 		err = z_erofs_extent_lookback(&m, m.delta[0]);
650 		if (err)
651 			goto unmap_out;
652 		break;
653 	default:
654 		erofs_err(inode->i_sb,
655 			  "unknown type %u @ offset %llu of nid %llu",
656 			  m.type, ofs, vi->nid);
657 		err = -EOPNOTSUPP;
658 		goto unmap_out;
659 	}
660 
661 	map->m_llen = end - map->m_la;
662 	map->m_pa = blknr_to_addr(m.pblk);
663 
664 	err = z_erofs_get_extent_compressedlen(&m, initial_lcn);
665 	if (err)
666 		goto out;
667 
668 	if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN)
669 		map->m_algorithmformat = Z_EROFS_COMPRESSION_SHIFTED;
670 	else if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2)
671 		map->m_algorithmformat = vi->z_algorithmtype[1];
672 	else
673 		map->m_algorithmformat = vi->z_algorithmtype[0];
674 
675 	if ((flags & EROFS_GET_BLOCKS_FIEMAP) ||
676 	    ((flags & EROFS_GET_BLOCKS_READMORE) &&
677 	     map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA &&
678 	     map->m_llen >= EROFS_BLKSIZ)) {
679 		err = z_erofs_get_extent_decompressedlen(&m);
680 		if (!err)
681 			map->m_flags |= EROFS_MAP_FULL_MAPPED;
682 	}
683 unmap_out:
684 	if (m.kaddr)
685 		kunmap_atomic(m.kaddr);
686 
687 out:
688 	erofs_dbg("%s, m_la %llu m_pa %llu m_llen %llu m_plen %llu m_flags 0%o",
689 		  __func__, map->m_la, map->m_pa,
690 		  map->m_llen, map->m_plen, map->m_flags);
691 
692 	trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
693 
694 	/* aggressively BUG_ON iff CONFIG_EROFS_FS_DEBUG is on */
695 	DBG_BUGON(err < 0 && err != -ENOMEM);
696 	return err;
697 }
698 
699 static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
700 				loff_t length, unsigned int flags,
701 				struct iomap *iomap, struct iomap *srcmap)
702 {
703 	int ret;
704 	struct erofs_map_blocks map = { .m_la = offset };
705 
706 	ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
707 	if (map.mpage)
708 		put_page(map.mpage);
709 	if (ret < 0)
710 		return ret;
711 
712 	iomap->bdev = inode->i_sb->s_bdev;
713 	iomap->offset = map.m_la;
714 	iomap->length = map.m_llen;
715 	if (map.m_flags & EROFS_MAP_MAPPED) {
716 		iomap->type = IOMAP_MAPPED;
717 		iomap->addr = map.m_pa;
718 	} else {
719 		iomap->type = IOMAP_HOLE;
720 		iomap->addr = IOMAP_NULL_ADDR;
721 		/*
722 		 * No strict rule how to describe extents for post EOF, yet
723 		 * we need do like below. Otherwise, iomap itself will get
724 		 * into an endless loop on post EOF.
725 		 */
726 		if (iomap->offset >= inode->i_size)
727 			iomap->length = length + map.m_la - offset;
728 	}
729 	iomap->flags = 0;
730 	return 0;
731 }
732 
733 const struct iomap_ops z_erofs_iomap_report_ops = {
734 	.iomap_begin = z_erofs_iomap_begin_report,
735 };
736