xref: /openbmc/linux/fs/erofs/zmap.c (revision a90fa0ad)
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) && !erofs_sb_has_fragments(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 	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
59 		    vi->xattr_isize, 8);
60 	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(pos), EROFS_KMAP);
61 	if (IS_ERR(kaddr)) {
62 		err = PTR_ERR(kaddr);
63 		goto out_unlock;
64 	}
65 
66 	h = kaddr + erofs_blkoff(pos);
67 	/*
68 	 * if the highest bit of the 8-byte map header is set, the whole file
69 	 * is stored in the packed inode. The rest bits keeps z_fragmentoff.
70 	 */
71 	if (h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT) {
72 		vi->z_advise = Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
73 		vi->z_fragmentoff = le64_to_cpu(*(__le64 *)h) ^ (1ULL << 63);
74 		vi->z_tailextent_headlcn = 0;
75 		goto done;
76 	}
77 	vi->z_advise = le16_to_cpu(h->h_advise);
78 	vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
79 	vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
80 
81 	headnr = 0;
82 	if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX ||
83 	    vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) {
84 		erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel",
85 			  headnr + 1, vi->z_algorithmtype[headnr], vi->nid);
86 		err = -EOPNOTSUPP;
87 		goto out_put_metabuf;
88 	}
89 
90 	vi->z_logical_clusterbits = LOG_BLOCK_SIZE + (h->h_clusterbits & 7);
91 	if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
92 	    vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
93 			    Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
94 		erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu",
95 			  vi->nid);
96 		err = -EFSCORRUPTED;
97 		goto out_put_metabuf;
98 	}
99 	if (vi->datalayout == EROFS_INODE_FLAT_COMPRESSION &&
100 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^
101 	    !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
102 		erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu",
103 			  vi->nid);
104 		err = -EFSCORRUPTED;
105 		goto out_put_metabuf;
106 	}
107 
108 	if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
109 		struct erofs_map_blocks map = {
110 			.buf = __EROFS_BUF_INITIALIZER
111 		};
112 
113 		vi->z_idata_size = le16_to_cpu(h->h_idata_size);
114 		err = z_erofs_do_map_blocks(inode, &map,
115 					    EROFS_GET_BLOCKS_FINDTAIL);
116 		erofs_put_metabuf(&map.buf);
117 
118 		if (!map.m_plen ||
119 		    erofs_blkoff(map.m_pa) + map.m_plen > EROFS_BLKSIZ) {
120 			erofs_err(sb, "invalid tail-packing pclustersize %llu",
121 				  map.m_plen);
122 			err = -EFSCORRUPTED;
123 		}
124 		if (err < 0)
125 			goto out_put_metabuf;
126 	}
127 
128 	if (vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER &&
129 	    !(h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT)) {
130 		struct erofs_map_blocks map = {
131 			.buf = __EROFS_BUF_INITIALIZER
132 		};
133 
134 		vi->z_fragmentoff = le32_to_cpu(h->h_fragmentoff);
135 		err = z_erofs_do_map_blocks(inode, &map,
136 					    EROFS_GET_BLOCKS_FINDTAIL);
137 		erofs_put_metabuf(&map.buf);
138 		if (err < 0)
139 			goto out_put_metabuf;
140 	}
141 done:
142 	/* paired with smp_mb() at the beginning of the function */
143 	smp_mb();
144 	set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
145 out_put_metabuf:
146 	erofs_put_metabuf(&buf);
147 out_unlock:
148 	clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
149 	return err;
150 }
151 
152 struct z_erofs_maprecorder {
153 	struct inode *inode;
154 	struct erofs_map_blocks *map;
155 	void *kaddr;
156 
157 	unsigned long lcn;
158 	/* compression extent information gathered */
159 	u8  type, headtype;
160 	u16 clusterofs;
161 	u16 delta[2];
162 	erofs_blk_t pblk, compressedblks;
163 	erofs_off_t nextpackoff;
164 	bool partialref;
165 };
166 
167 static int legacy_load_cluster_from_disk(struct z_erofs_maprecorder *m,
168 					 unsigned long lcn)
169 {
170 	struct inode *const inode = m->inode;
171 	struct erofs_inode *const vi = EROFS_I(inode);
172 	const erofs_off_t ibase = iloc(EROFS_I_SB(inode), vi->nid);
173 	const erofs_off_t pos =
174 		Z_EROFS_VLE_LEGACY_INDEX_ALIGN(ibase + vi->inode_isize +
175 					       vi->xattr_isize) +
176 		lcn * sizeof(struct z_erofs_vle_decompressed_index);
177 	struct z_erofs_vle_decompressed_index *di;
178 	unsigned int advise, type;
179 
180 	m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
181 				      erofs_blknr(pos), EROFS_KMAP);
182 	if (IS_ERR(m->kaddr))
183 		return PTR_ERR(m->kaddr);
184 
185 	m->nextpackoff = pos + sizeof(struct z_erofs_vle_decompressed_index);
186 	m->lcn = lcn;
187 	di = m->kaddr + erofs_blkoff(pos);
188 
189 	advise = le16_to_cpu(di->di_advise);
190 	type = (advise >> Z_EROFS_VLE_DI_CLUSTER_TYPE_BIT) &
191 		((1 << Z_EROFS_VLE_DI_CLUSTER_TYPE_BITS) - 1);
192 	switch (type) {
193 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
194 		m->clusterofs = 1 << vi->z_logical_clusterbits;
195 		m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
196 		if (m->delta[0] & Z_EROFS_VLE_DI_D0_CBLKCNT) {
197 			if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
198 					Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
199 				DBG_BUGON(1);
200 				return -EFSCORRUPTED;
201 			}
202 			m->compressedblks = m->delta[0] &
203 				~Z_EROFS_VLE_DI_D0_CBLKCNT;
204 			m->delta[0] = 1;
205 		}
206 		m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
207 		break;
208 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
209 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
210 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
211 		if (advise & Z_EROFS_VLE_DI_PARTIAL_REF)
212 			m->partialref = true;
213 		m->clusterofs = le16_to_cpu(di->di_clusterofs);
214 		m->pblk = le32_to_cpu(di->di_u.blkaddr);
215 		break;
216 	default:
217 		DBG_BUGON(1);
218 		return -EOPNOTSUPP;
219 	}
220 	m->type = type;
221 	return 0;
222 }
223 
224 static unsigned int decode_compactedbits(unsigned int lobits,
225 					 unsigned int lomask,
226 					 u8 *in, unsigned int pos, u8 *type)
227 {
228 	const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
229 	const unsigned int lo = v & lomask;
230 
231 	*type = (v >> lobits) & 3;
232 	return lo;
233 }
234 
235 static int get_compacted_la_distance(unsigned int lclusterbits,
236 				     unsigned int encodebits,
237 				     unsigned int vcnt, u8 *in, int i)
238 {
239 	const unsigned int lomask = (1 << lclusterbits) - 1;
240 	unsigned int lo, d1 = 0;
241 	u8 type;
242 
243 	DBG_BUGON(i >= vcnt);
244 
245 	do {
246 		lo = decode_compactedbits(lclusterbits, lomask,
247 					  in, encodebits * i, &type);
248 
249 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
250 			return d1;
251 		++d1;
252 	} while (++i < vcnt);
253 
254 	/* vcnt - 1 (Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) item */
255 	if (!(lo & Z_EROFS_VLE_DI_D0_CBLKCNT))
256 		d1 += lo - 1;
257 	return d1;
258 }
259 
260 static int unpack_compacted_index(struct z_erofs_maprecorder *m,
261 				  unsigned int amortizedshift,
262 				  erofs_off_t pos, bool lookahead)
263 {
264 	struct erofs_inode *const vi = EROFS_I(m->inode);
265 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
266 	const unsigned int lomask = (1 << lclusterbits) - 1;
267 	unsigned int vcnt, base, lo, encodebits, nblk, eofs;
268 	int i;
269 	u8 *in, type;
270 	bool big_pcluster;
271 
272 	if (1 << amortizedshift == 4)
273 		vcnt = 2;
274 	else if (1 << amortizedshift == 2 && lclusterbits == 12)
275 		vcnt = 16;
276 	else
277 		return -EOPNOTSUPP;
278 
279 	/* it doesn't equal to round_up(..) */
280 	m->nextpackoff = round_down(pos, vcnt << amortizedshift) +
281 			 (vcnt << amortizedshift);
282 	big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1;
283 	encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
284 	eofs = erofs_blkoff(pos);
285 	base = round_down(eofs, vcnt << amortizedshift);
286 	in = m->kaddr + base;
287 
288 	i = (eofs - base) >> amortizedshift;
289 
290 	lo = decode_compactedbits(lclusterbits, lomask,
291 				  in, encodebits * i, &type);
292 	m->type = type;
293 	if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
294 		m->clusterofs = 1 << lclusterbits;
295 
296 		/* figure out lookahead_distance: delta[1] if needed */
297 		if (lookahead)
298 			m->delta[1] = get_compacted_la_distance(lclusterbits,
299 						encodebits, vcnt, in, i);
300 		if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
301 			if (!big_pcluster) {
302 				DBG_BUGON(1);
303 				return -EFSCORRUPTED;
304 			}
305 			m->compressedblks = lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
306 			m->delta[0] = 1;
307 			return 0;
308 		} else if (i + 1 != (int)vcnt) {
309 			m->delta[0] = lo;
310 			return 0;
311 		}
312 		/*
313 		 * since the last lcluster in the pack is special,
314 		 * of which lo saves delta[1] rather than delta[0].
315 		 * Hence, get delta[0] by the previous lcluster indirectly.
316 		 */
317 		lo = decode_compactedbits(lclusterbits, lomask,
318 					  in, encodebits * (i - 1), &type);
319 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
320 			lo = 0;
321 		else if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT)
322 			lo = 1;
323 		m->delta[0] = lo + 1;
324 		return 0;
325 	}
326 	m->clusterofs = lo;
327 	m->delta[0] = 0;
328 	/* figout out blkaddr (pblk) for HEAD lclusters */
329 	if (!big_pcluster) {
330 		nblk = 1;
331 		while (i > 0) {
332 			--i;
333 			lo = decode_compactedbits(lclusterbits, lomask,
334 						  in, encodebits * i, &type);
335 			if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
336 				i -= lo;
337 
338 			if (i >= 0)
339 				++nblk;
340 		}
341 	} else {
342 		nblk = 0;
343 		while (i > 0) {
344 			--i;
345 			lo = decode_compactedbits(lclusterbits, lomask,
346 						  in, encodebits * i, &type);
347 			if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
348 				if (lo & Z_EROFS_VLE_DI_D0_CBLKCNT) {
349 					--i;
350 					nblk += lo & ~Z_EROFS_VLE_DI_D0_CBLKCNT;
351 					continue;
352 				}
353 				/* bigpcluster shouldn't have plain d0 == 1 */
354 				if (lo <= 1) {
355 					DBG_BUGON(1);
356 					return -EFSCORRUPTED;
357 				}
358 				i -= lo - 2;
359 				continue;
360 			}
361 			++nblk;
362 		}
363 	}
364 	in += (vcnt << amortizedshift) - sizeof(__le32);
365 	m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
366 	return 0;
367 }
368 
369 static int compacted_load_cluster_from_disk(struct z_erofs_maprecorder *m,
370 					    unsigned long lcn, bool lookahead)
371 {
372 	struct inode *const inode = m->inode;
373 	struct erofs_inode *const vi = EROFS_I(inode);
374 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
375 	const erofs_off_t ebase = ALIGN(iloc(EROFS_I_SB(inode), vi->nid) +
376 					vi->inode_isize + vi->xattr_isize, 8) +
377 		sizeof(struct z_erofs_map_header);
378 	const unsigned int totalidx = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
379 	unsigned int compacted_4b_initial, compacted_2b;
380 	unsigned int amortizedshift;
381 	erofs_off_t pos;
382 
383 	if (lclusterbits != 12)
384 		return -EOPNOTSUPP;
385 
386 	if (lcn >= totalidx)
387 		return -EINVAL;
388 
389 	m->lcn = lcn;
390 	/* used to align to 32-byte (compacted_2b) alignment */
391 	compacted_4b_initial = (32 - ebase % 32) / 4;
392 	if (compacted_4b_initial == 32 / 4)
393 		compacted_4b_initial = 0;
394 
395 	if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
396 	    compacted_4b_initial < totalidx)
397 		compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
398 	else
399 		compacted_2b = 0;
400 
401 	pos = ebase;
402 	if (lcn < compacted_4b_initial) {
403 		amortizedshift = 2;
404 		goto out;
405 	}
406 	pos += compacted_4b_initial * 4;
407 	lcn -= compacted_4b_initial;
408 
409 	if (lcn < compacted_2b) {
410 		amortizedshift = 1;
411 		goto out;
412 	}
413 	pos += compacted_2b * 2;
414 	lcn -= compacted_2b;
415 	amortizedshift = 2;
416 out:
417 	pos += lcn * (1 << amortizedshift);
418 	m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
419 				      erofs_blknr(pos), EROFS_KMAP);
420 	if (IS_ERR(m->kaddr))
421 		return PTR_ERR(m->kaddr);
422 	return unpack_compacted_index(m, amortizedshift, pos, lookahead);
423 }
424 
425 static int z_erofs_load_cluster_from_disk(struct z_erofs_maprecorder *m,
426 					  unsigned int lcn, bool lookahead)
427 {
428 	const unsigned int datamode = EROFS_I(m->inode)->datalayout;
429 
430 	if (datamode == EROFS_INODE_FLAT_COMPRESSION_LEGACY)
431 		return legacy_load_cluster_from_disk(m, lcn);
432 
433 	if (datamode == EROFS_INODE_FLAT_COMPRESSION)
434 		return compacted_load_cluster_from_disk(m, lcn, lookahead);
435 
436 	return -EINVAL;
437 }
438 
439 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
440 				   unsigned int lookback_distance)
441 {
442 	struct erofs_inode *const vi = EROFS_I(m->inode);
443 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
444 
445 	while (m->lcn >= lookback_distance) {
446 		unsigned long lcn = m->lcn - lookback_distance;
447 		int err;
448 
449 		/* load extent head logical cluster if needed */
450 		err = z_erofs_load_cluster_from_disk(m, lcn, false);
451 		if (err)
452 			return err;
453 
454 		switch (m->type) {
455 		case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
456 			if (!m->delta[0]) {
457 				erofs_err(m->inode->i_sb,
458 					  "invalid lookback distance 0 @ nid %llu",
459 					  vi->nid);
460 				DBG_BUGON(1);
461 				return -EFSCORRUPTED;
462 			}
463 			lookback_distance = m->delta[0];
464 			continue;
465 		case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
466 		case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
467 		case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
468 			m->headtype = m->type;
469 			m->map->m_la = (lcn << lclusterbits) | m->clusterofs;
470 			return 0;
471 		default:
472 			erofs_err(m->inode->i_sb,
473 				  "unknown type %u @ lcn %lu of nid %llu",
474 				  m->type, lcn, vi->nid);
475 			DBG_BUGON(1);
476 			return -EOPNOTSUPP;
477 		}
478 	}
479 
480 	erofs_err(m->inode->i_sb, "bogus lookback distance @ nid %llu",
481 		  vi->nid);
482 	DBG_BUGON(1);
483 	return -EFSCORRUPTED;
484 }
485 
486 static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m,
487 					    unsigned int initial_lcn)
488 {
489 	struct erofs_inode *const vi = EROFS_I(m->inode);
490 	struct erofs_map_blocks *const map = m->map;
491 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
492 	unsigned long lcn;
493 	int err;
494 
495 	DBG_BUGON(m->type != Z_EROFS_VLE_CLUSTER_TYPE_PLAIN &&
496 		  m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 &&
497 		  m->type != Z_EROFS_VLE_CLUSTER_TYPE_HEAD2);
498 	DBG_BUGON(m->type != m->headtype);
499 
500 	if (m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
501 	    ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1) &&
502 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1)) ||
503 	    ((m->headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) &&
504 	     !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
505 		map->m_plen = 1ULL << lclusterbits;
506 		return 0;
507 	}
508 	lcn = m->lcn + 1;
509 	if (m->compressedblks)
510 		goto out;
511 
512 	err = z_erofs_load_cluster_from_disk(m, lcn, false);
513 	if (err)
514 		return err;
515 
516 	/*
517 	 * If the 1st NONHEAD lcluster has already been handled initially w/o
518 	 * valid compressedblks, which means at least it mustn't be CBLKCNT, or
519 	 * an internal implemenatation error is detected.
520 	 *
521 	 * The following code can also handle it properly anyway, but let's
522 	 * BUG_ON in the debugging mode only for developers to notice that.
523 	 */
524 	DBG_BUGON(lcn == initial_lcn &&
525 		  m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD);
526 
527 	switch (m->type) {
528 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
529 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
530 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
531 		/*
532 		 * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type
533 		 * rather than CBLKCNT, it's a 1 lcluster-sized pcluster.
534 		 */
535 		m->compressedblks = 1 << (lclusterbits - LOG_BLOCK_SIZE);
536 		break;
537 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
538 		if (m->delta[0] != 1)
539 			goto err_bonus_cblkcnt;
540 		if (m->compressedblks)
541 			break;
542 		fallthrough;
543 	default:
544 		erofs_err(m->inode->i_sb,
545 			  "cannot found CBLKCNT @ lcn %lu of nid %llu",
546 			  lcn, vi->nid);
547 		DBG_BUGON(1);
548 		return -EFSCORRUPTED;
549 	}
550 out:
551 	map->m_plen = (u64)m->compressedblks << LOG_BLOCK_SIZE;
552 	return 0;
553 err_bonus_cblkcnt:
554 	erofs_err(m->inode->i_sb,
555 		  "bogus CBLKCNT @ lcn %lu of nid %llu",
556 		  lcn, vi->nid);
557 	DBG_BUGON(1);
558 	return -EFSCORRUPTED;
559 }
560 
561 static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m)
562 {
563 	struct inode *inode = m->inode;
564 	struct erofs_inode *vi = EROFS_I(inode);
565 	struct erofs_map_blocks *map = m->map;
566 	unsigned int lclusterbits = vi->z_logical_clusterbits;
567 	u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits;
568 	int err;
569 
570 	do {
571 		/* handle the last EOF pcluster (no next HEAD lcluster) */
572 		if ((lcn << lclusterbits) >= inode->i_size) {
573 			map->m_llen = inode->i_size - map->m_la;
574 			return 0;
575 		}
576 
577 		err = z_erofs_load_cluster_from_disk(m, lcn, true);
578 		if (err)
579 			return err;
580 
581 		if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
582 			DBG_BUGON(!m->delta[1] &&
583 				  m->clusterofs != 1 << lclusterbits);
584 		} else if (m->type == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN ||
585 			   m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD1 ||
586 			   m->type == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) {
587 			/* go on until the next HEAD lcluster */
588 			if (lcn != headlcn)
589 				break;
590 			m->delta[1] = 1;
591 		} else {
592 			erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu",
593 				  m->type, lcn, vi->nid);
594 			DBG_BUGON(1);
595 			return -EOPNOTSUPP;
596 		}
597 		lcn += m->delta[1];
598 	} while (m->delta[1]);
599 
600 	map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la;
601 	return 0;
602 }
603 
604 static int z_erofs_do_map_blocks(struct inode *inode,
605 				 struct erofs_map_blocks *map,
606 				 int flags)
607 {
608 	struct erofs_inode *const vi = EROFS_I(inode);
609 	bool ztailpacking = vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER;
610 	bool fragment = vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
611 	struct z_erofs_maprecorder m = {
612 		.inode = inode,
613 		.map = map,
614 	};
615 	int err = 0;
616 	unsigned int lclusterbits, endoff;
617 	unsigned long initial_lcn;
618 	unsigned long long ofs, end;
619 
620 	lclusterbits = vi->z_logical_clusterbits;
621 	ofs = flags & EROFS_GET_BLOCKS_FINDTAIL ? inode->i_size - 1 : map->m_la;
622 	initial_lcn = ofs >> lclusterbits;
623 	endoff = ofs & ((1 << lclusterbits) - 1);
624 
625 	err = z_erofs_load_cluster_from_disk(&m, initial_lcn, false);
626 	if (err)
627 		goto unmap_out;
628 
629 	if (ztailpacking && (flags & EROFS_GET_BLOCKS_FINDTAIL))
630 		vi->z_idataoff = m.nextpackoff;
631 
632 	map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED;
633 	end = (m.lcn + 1ULL) << lclusterbits;
634 
635 	switch (m.type) {
636 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
637 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD1:
638 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD2:
639 		if (endoff >= m.clusterofs) {
640 			m.headtype = m.type;
641 			map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
642 			/*
643 			 * For ztailpacking files, in order to inline data more
644 			 * effectively, special EOF lclusters are now supported
645 			 * which can have three parts at most.
646 			 */
647 			if (ztailpacking && end > inode->i_size)
648 				end = inode->i_size;
649 			break;
650 		}
651 		/* m.lcn should be >= 1 if endoff < m.clusterofs */
652 		if (!m.lcn) {
653 			erofs_err(inode->i_sb,
654 				  "invalid logical cluster 0 at nid %llu",
655 				  vi->nid);
656 			err = -EFSCORRUPTED;
657 			goto unmap_out;
658 		}
659 		end = (m.lcn << lclusterbits) | m.clusterofs;
660 		map->m_flags |= EROFS_MAP_FULL_MAPPED;
661 		m.delta[0] = 1;
662 		fallthrough;
663 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
664 		/* get the corresponding first chunk */
665 		err = z_erofs_extent_lookback(&m, m.delta[0]);
666 		if (err)
667 			goto unmap_out;
668 		break;
669 	default:
670 		erofs_err(inode->i_sb,
671 			  "unknown type %u @ offset %llu of nid %llu",
672 			  m.type, ofs, vi->nid);
673 		err = -EOPNOTSUPP;
674 		goto unmap_out;
675 	}
676 	if (m.partialref)
677 		map->m_flags |= EROFS_MAP_PARTIAL_REF;
678 	map->m_llen = end - map->m_la;
679 
680 	if (flags & EROFS_GET_BLOCKS_FINDTAIL) {
681 		vi->z_tailextent_headlcn = m.lcn;
682 		/* for non-compact indexes, fragmentoff is 64 bits */
683 		if (fragment &&
684 		    vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY)
685 			vi->z_fragmentoff |= (u64)m.pblk << 32;
686 	}
687 	if (ztailpacking && m.lcn == vi->z_tailextent_headlcn) {
688 		map->m_flags |= EROFS_MAP_META;
689 		map->m_pa = vi->z_idataoff;
690 		map->m_plen = vi->z_idata_size;
691 	} else if (fragment && m.lcn == vi->z_tailextent_headlcn) {
692 		map->m_flags |= EROFS_MAP_FRAGMENT;
693 	} else {
694 		map->m_pa = blknr_to_addr(m.pblk);
695 		err = z_erofs_get_extent_compressedlen(&m, initial_lcn);
696 		if (err)
697 			goto unmap_out;
698 	}
699 
700 	if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_PLAIN) {
701 		if (map->m_llen > map->m_plen) {
702 			DBG_BUGON(1);
703 			err = -EFSCORRUPTED;
704 			goto unmap_out;
705 		}
706 		if (vi->z_advise & Z_EROFS_ADVISE_INTERLACED_PCLUSTER)
707 			map->m_algorithmformat =
708 				Z_EROFS_COMPRESSION_INTERLACED;
709 		else
710 			map->m_algorithmformat =
711 				Z_EROFS_COMPRESSION_SHIFTED;
712 	} else if (m.headtype == Z_EROFS_VLE_CLUSTER_TYPE_HEAD2) {
713 		map->m_algorithmformat = vi->z_algorithmtype[1];
714 	} else {
715 		map->m_algorithmformat = vi->z_algorithmtype[0];
716 	}
717 
718 	if ((flags & EROFS_GET_BLOCKS_FIEMAP) ||
719 	    ((flags & EROFS_GET_BLOCKS_READMORE) &&
720 	     map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA &&
721 	     map->m_llen >= EROFS_BLKSIZ)) {
722 		err = z_erofs_get_extent_decompressedlen(&m);
723 		if (!err)
724 			map->m_flags |= EROFS_MAP_FULL_MAPPED;
725 	}
726 
727 unmap_out:
728 	erofs_unmap_metabuf(&m.map->buf);
729 	erofs_dbg("%s, m_la %llu m_pa %llu m_llen %llu m_plen %llu m_flags 0%o",
730 		  __func__, map->m_la, map->m_pa,
731 		  map->m_llen, map->m_plen, map->m_flags);
732 	return err;
733 }
734 
735 int z_erofs_map_blocks_iter(struct inode *inode, struct erofs_map_blocks *map,
736 			    int flags)
737 {
738 	struct erofs_inode *const vi = EROFS_I(inode);
739 	int err = 0;
740 
741 	trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
742 
743 	/* when trying to read beyond EOF, leave it unmapped */
744 	if (map->m_la >= inode->i_size) {
745 		map->m_llen = map->m_la + 1 - inode->i_size;
746 		map->m_la = inode->i_size;
747 		map->m_flags = 0;
748 		goto out;
749 	}
750 
751 	err = z_erofs_fill_inode_lazy(inode);
752 	if (err)
753 		goto out;
754 
755 	if ((vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER) &&
756 	    !vi->z_tailextent_headlcn) {
757 		map->m_la = 0;
758 		map->m_llen = inode->i_size;
759 		map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_FULL_MAPPED |
760 				EROFS_MAP_FRAGMENT;
761 		goto out;
762 	}
763 
764 	err = z_erofs_do_map_blocks(inode, map, flags);
765 out:
766 	trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
767 
768 	/* aggressively BUG_ON iff CONFIG_EROFS_FS_DEBUG is on */
769 	DBG_BUGON(err < 0 && err != -ENOMEM);
770 	return err;
771 }
772 
773 static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
774 				loff_t length, unsigned int flags,
775 				struct iomap *iomap, struct iomap *srcmap)
776 {
777 	int ret;
778 	struct erofs_map_blocks map = { .m_la = offset };
779 
780 	ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
781 	erofs_put_metabuf(&map.buf);
782 	if (ret < 0)
783 		return ret;
784 
785 	iomap->bdev = inode->i_sb->s_bdev;
786 	iomap->offset = map.m_la;
787 	iomap->length = map.m_llen;
788 	if (map.m_flags & EROFS_MAP_MAPPED) {
789 		iomap->type = IOMAP_MAPPED;
790 		iomap->addr = map.m_flags & EROFS_MAP_FRAGMENT ?
791 			      IOMAP_NULL_ADDR : map.m_pa;
792 	} else {
793 		iomap->type = IOMAP_HOLE;
794 		iomap->addr = IOMAP_NULL_ADDR;
795 		/*
796 		 * No strict rule on how to describe extents for post EOF, yet
797 		 * we need to do like below. Otherwise, iomap itself will get
798 		 * into an endless loop on post EOF.
799 		 *
800 		 * Calculate the effective offset by subtracting extent start
801 		 * (map.m_la) from the requested offset, and add it to length.
802 		 * (NB: offset >= map.m_la always)
803 		 */
804 		if (iomap->offset >= inode->i_size)
805 			iomap->length = length + offset - map.m_la;
806 	}
807 	iomap->flags = 0;
808 	return 0;
809 }
810 
811 const struct iomap_ops z_erofs_iomap_report_ops = {
812 	.iomap_begin = z_erofs_iomap_begin_report,
813 };
814