Loading...
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
10struct z_erofs_maprecorder {
11 struct inode *inode;
12 struct erofs_map_blocks *map;
13 void *kaddr;
14
15 unsigned long lcn;
16 /* compression extent information gathered */
17 u8 type, headtype;
18 u16 clusterofs;
19 u16 delta[2];
20 erofs_blk_t pblk, compressedblks;
21 erofs_off_t nextpackoff;
22 bool partialref;
23};
24
25static int z_erofs_load_full_lcluster(struct z_erofs_maprecorder *m,
26 unsigned long lcn)
27{
28 struct inode *const inode = m->inode;
29 struct erofs_inode *const vi = EROFS_I(inode);
30 const erofs_off_t pos = Z_EROFS_FULL_INDEX_ALIGN(erofs_iloc(inode) +
31 vi->inode_isize + vi->xattr_isize) +
32 lcn * sizeof(struct z_erofs_lcluster_index);
33 struct z_erofs_lcluster_index *di;
34 unsigned int advise, type;
35
36 m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
37 erofs_blknr(inode->i_sb, pos), EROFS_KMAP);
38 if (IS_ERR(m->kaddr))
39 return PTR_ERR(m->kaddr);
40
41 m->nextpackoff = pos + sizeof(struct z_erofs_lcluster_index);
42 m->lcn = lcn;
43 di = m->kaddr + erofs_blkoff(inode->i_sb, pos);
44
45 advise = le16_to_cpu(di->di_advise);
46 type = (advise >> Z_EROFS_LI_LCLUSTER_TYPE_BIT) &
47 ((1 << Z_EROFS_LI_LCLUSTER_TYPE_BITS) - 1);
48 switch (type) {
49 case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
50 m->clusterofs = 1 << vi->z_logical_clusterbits;
51 m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
52 if (m->delta[0] & Z_EROFS_LI_D0_CBLKCNT) {
53 if (!(vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
54 Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
55 DBG_BUGON(1);
56 return -EFSCORRUPTED;
57 }
58 m->compressedblks = m->delta[0] &
59 ~Z_EROFS_LI_D0_CBLKCNT;
60 m->delta[0] = 1;
61 }
62 m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
63 break;
64 case Z_EROFS_LCLUSTER_TYPE_PLAIN:
65 case Z_EROFS_LCLUSTER_TYPE_HEAD1:
66 case Z_EROFS_LCLUSTER_TYPE_HEAD2:
67 if (advise & Z_EROFS_LI_PARTIAL_REF)
68 m->partialref = true;
69 m->clusterofs = le16_to_cpu(di->di_clusterofs);
70 if (m->clusterofs >= 1 << vi->z_logical_clusterbits) {
71 DBG_BUGON(1);
72 return -EFSCORRUPTED;
73 }
74 m->pblk = le32_to_cpu(di->di_u.blkaddr);
75 break;
76 default:
77 DBG_BUGON(1);
78 return -EOPNOTSUPP;
79 }
80 m->type = type;
81 return 0;
82}
83
84static unsigned int decode_compactedbits(unsigned int lobits,
85 u8 *in, unsigned int pos, u8 *type)
86{
87 const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
88 const unsigned int lo = v & ((1 << lobits) - 1);
89
90 *type = (v >> lobits) & 3;
91 return lo;
92}
93
94static int get_compacted_la_distance(unsigned int lobits,
95 unsigned int encodebits,
96 unsigned int vcnt, u8 *in, int i)
97{
98 unsigned int lo, d1 = 0;
99 u8 type;
100
101 DBG_BUGON(i >= vcnt);
102
103 do {
104 lo = decode_compactedbits(lobits, in, encodebits * i, &type);
105
106 if (type != Z_EROFS_LCLUSTER_TYPE_NONHEAD)
107 return d1;
108 ++d1;
109 } while (++i < vcnt);
110
111 /* vcnt - 1 (Z_EROFS_LCLUSTER_TYPE_NONHEAD) item */
112 if (!(lo & Z_EROFS_LI_D0_CBLKCNT))
113 d1 += lo - 1;
114 return d1;
115}
116
117static int unpack_compacted_index(struct z_erofs_maprecorder *m,
118 unsigned int amortizedshift,
119 erofs_off_t pos, bool lookahead)
120{
121 struct erofs_inode *const vi = EROFS_I(m->inode);
122 const unsigned int lclusterbits = vi->z_logical_clusterbits;
123 unsigned int vcnt, base, lo, lobits, encodebits, nblk, eofs;
124 int i;
125 u8 *in, type;
126 bool big_pcluster;
127
128 if (1 << amortizedshift == 4 && lclusterbits <= 14)
129 vcnt = 2;
130 else if (1 << amortizedshift == 2 && lclusterbits <= 12)
131 vcnt = 16;
132 else
133 return -EOPNOTSUPP;
134
135 /* it doesn't equal to round_up(..) */
136 m->nextpackoff = round_down(pos, vcnt << amortizedshift) +
137 (vcnt << amortizedshift);
138 big_pcluster = vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1;
139 lobits = max(lclusterbits, ilog2(Z_EROFS_LI_D0_CBLKCNT) + 1U);
140 encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
141 eofs = erofs_blkoff(m->inode->i_sb, pos);
142 base = round_down(eofs, vcnt << amortizedshift);
143 in = m->kaddr + base;
144
145 i = (eofs - base) >> amortizedshift;
146
147 lo = decode_compactedbits(lobits, in, encodebits * i, &type);
148 m->type = type;
149 if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
150 m->clusterofs = 1 << lclusterbits;
151
152 /* figure out lookahead_distance: delta[1] if needed */
153 if (lookahead)
154 m->delta[1] = get_compacted_la_distance(lobits,
155 encodebits, vcnt, in, i);
156 if (lo & Z_EROFS_LI_D0_CBLKCNT) {
157 if (!big_pcluster) {
158 DBG_BUGON(1);
159 return -EFSCORRUPTED;
160 }
161 m->compressedblks = lo & ~Z_EROFS_LI_D0_CBLKCNT;
162 m->delta[0] = 1;
163 return 0;
164 } else if (i + 1 != (int)vcnt) {
165 m->delta[0] = lo;
166 return 0;
167 }
168 /*
169 * since the last lcluster in the pack is special,
170 * of which lo saves delta[1] rather than delta[0].
171 * Hence, get delta[0] by the previous lcluster indirectly.
172 */
173 lo = decode_compactedbits(lobits, in,
174 encodebits * (i - 1), &type);
175 if (type != Z_EROFS_LCLUSTER_TYPE_NONHEAD)
176 lo = 0;
177 else if (lo & Z_EROFS_LI_D0_CBLKCNT)
178 lo = 1;
179 m->delta[0] = lo + 1;
180 return 0;
181 }
182 m->clusterofs = lo;
183 m->delta[0] = 0;
184 /* figout out blkaddr (pblk) for HEAD lclusters */
185 if (!big_pcluster) {
186 nblk = 1;
187 while (i > 0) {
188 --i;
189 lo = decode_compactedbits(lobits, in,
190 encodebits * i, &type);
191 if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD)
192 i -= lo;
193
194 if (i >= 0)
195 ++nblk;
196 }
197 } else {
198 nblk = 0;
199 while (i > 0) {
200 --i;
201 lo = decode_compactedbits(lobits, in,
202 encodebits * i, &type);
203 if (type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
204 if (lo & Z_EROFS_LI_D0_CBLKCNT) {
205 --i;
206 nblk += lo & ~Z_EROFS_LI_D0_CBLKCNT;
207 continue;
208 }
209 /* bigpcluster shouldn't have plain d0 == 1 */
210 if (lo <= 1) {
211 DBG_BUGON(1);
212 return -EFSCORRUPTED;
213 }
214 i -= lo - 2;
215 continue;
216 }
217 ++nblk;
218 }
219 }
220 in += (vcnt << amortizedshift) - sizeof(__le32);
221 m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
222 return 0;
223}
224
225static int z_erofs_load_compact_lcluster(struct z_erofs_maprecorder *m,
226 unsigned long lcn, bool lookahead)
227{
228 struct inode *const inode = m->inode;
229 struct erofs_inode *const vi = EROFS_I(inode);
230 const erofs_off_t ebase = sizeof(struct z_erofs_map_header) +
231 ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8);
232 unsigned int totalidx = erofs_iblks(inode);
233 unsigned int compacted_4b_initial, compacted_2b;
234 unsigned int amortizedshift;
235 erofs_off_t pos;
236
237 if (lcn >= totalidx)
238 return -EINVAL;
239
240 m->lcn = lcn;
241 /* used to align to 32-byte (compacted_2b) alignment */
242 compacted_4b_initial = (32 - ebase % 32) / 4;
243 if (compacted_4b_initial == 32 / 4)
244 compacted_4b_initial = 0;
245
246 if ((vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B) &&
247 compacted_4b_initial < totalidx)
248 compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
249 else
250 compacted_2b = 0;
251
252 pos = ebase;
253 if (lcn < compacted_4b_initial) {
254 amortizedshift = 2;
255 goto out;
256 }
257 pos += compacted_4b_initial * 4;
258 lcn -= compacted_4b_initial;
259
260 if (lcn < compacted_2b) {
261 amortizedshift = 1;
262 goto out;
263 }
264 pos += compacted_2b * 2;
265 lcn -= compacted_2b;
266 amortizedshift = 2;
267out:
268 pos += lcn * (1 << amortizedshift);
269 m->kaddr = erofs_read_metabuf(&m->map->buf, inode->i_sb,
270 erofs_blknr(inode->i_sb, pos), EROFS_KMAP);
271 if (IS_ERR(m->kaddr))
272 return PTR_ERR(m->kaddr);
273 return unpack_compacted_index(m, amortizedshift, pos, lookahead);
274}
275
276static int z_erofs_load_lcluster_from_disk(struct z_erofs_maprecorder *m,
277 unsigned int lcn, bool lookahead)
278{
279 switch (EROFS_I(m->inode)->datalayout) {
280 case EROFS_INODE_COMPRESSED_FULL:
281 return z_erofs_load_full_lcluster(m, lcn);
282 case EROFS_INODE_COMPRESSED_COMPACT:
283 return z_erofs_load_compact_lcluster(m, lcn, lookahead);
284 default:
285 return -EINVAL;
286 }
287}
288
289static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
290 unsigned int lookback_distance)
291{
292 struct super_block *sb = m->inode->i_sb;
293 struct erofs_inode *const vi = EROFS_I(m->inode);
294 const unsigned int lclusterbits = vi->z_logical_clusterbits;
295
296 while (m->lcn >= lookback_distance) {
297 unsigned long lcn = m->lcn - lookback_distance;
298 int err;
299
300 err = z_erofs_load_lcluster_from_disk(m, lcn, false);
301 if (err)
302 return err;
303
304 switch (m->type) {
305 case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
306 lookback_distance = m->delta[0];
307 if (!lookback_distance)
308 goto err_bogus;
309 continue;
310 case Z_EROFS_LCLUSTER_TYPE_PLAIN:
311 case Z_EROFS_LCLUSTER_TYPE_HEAD1:
312 case Z_EROFS_LCLUSTER_TYPE_HEAD2:
313 m->headtype = m->type;
314 m->map->m_la = (lcn << lclusterbits) | m->clusterofs;
315 return 0;
316 default:
317 erofs_err(sb, "unknown type %u @ lcn %lu of nid %llu",
318 m->type, lcn, vi->nid);
319 DBG_BUGON(1);
320 return -EOPNOTSUPP;
321 }
322 }
323err_bogus:
324 erofs_err(sb, "bogus lookback distance %u @ lcn %lu of nid %llu",
325 lookback_distance, m->lcn, vi->nid);
326 DBG_BUGON(1);
327 return -EFSCORRUPTED;
328}
329
330static int z_erofs_get_extent_compressedlen(struct z_erofs_maprecorder *m,
331 unsigned int initial_lcn)
332{
333 struct super_block *sb = m->inode->i_sb;
334 struct erofs_inode *const vi = EROFS_I(m->inode);
335 struct erofs_map_blocks *const map = m->map;
336 const unsigned int lclusterbits = vi->z_logical_clusterbits;
337 unsigned long lcn;
338 int err;
339
340 DBG_BUGON(m->type != Z_EROFS_LCLUSTER_TYPE_PLAIN &&
341 m->type != Z_EROFS_LCLUSTER_TYPE_HEAD1 &&
342 m->type != Z_EROFS_LCLUSTER_TYPE_HEAD2);
343 DBG_BUGON(m->type != m->headtype);
344
345 if (m->headtype == Z_EROFS_LCLUSTER_TYPE_PLAIN ||
346 ((m->headtype == Z_EROFS_LCLUSTER_TYPE_HEAD1) &&
347 !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1)) ||
348 ((m->headtype == Z_EROFS_LCLUSTER_TYPE_HEAD2) &&
349 !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2))) {
350 map->m_plen = 1ULL << lclusterbits;
351 return 0;
352 }
353 lcn = m->lcn + 1;
354 if (m->compressedblks)
355 goto out;
356
357 err = z_erofs_load_lcluster_from_disk(m, lcn, false);
358 if (err)
359 return err;
360
361 /*
362 * If the 1st NONHEAD lcluster has already been handled initially w/o
363 * valid compressedblks, which means at least it mustn't be CBLKCNT, or
364 * an internal implemenatation error is detected.
365 *
366 * The following code can also handle it properly anyway, but let's
367 * BUG_ON in the debugging mode only for developers to notice that.
368 */
369 DBG_BUGON(lcn == initial_lcn &&
370 m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD);
371
372 switch (m->type) {
373 case Z_EROFS_LCLUSTER_TYPE_PLAIN:
374 case Z_EROFS_LCLUSTER_TYPE_HEAD1:
375 case Z_EROFS_LCLUSTER_TYPE_HEAD2:
376 /*
377 * if the 1st NONHEAD lcluster is actually PLAIN or HEAD type
378 * rather than CBLKCNT, it's a 1 lcluster-sized pcluster.
379 */
380 m->compressedblks = 1 << (lclusterbits - sb->s_blocksize_bits);
381 break;
382 case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
383 if (m->delta[0] != 1)
384 goto err_bonus_cblkcnt;
385 if (m->compressedblks)
386 break;
387 fallthrough;
388 default:
389 erofs_err(sb, "cannot found CBLKCNT @ lcn %lu of nid %llu", lcn,
390 vi->nid);
391 DBG_BUGON(1);
392 return -EFSCORRUPTED;
393 }
394out:
395 map->m_plen = erofs_pos(sb, m->compressedblks);
396 return 0;
397err_bonus_cblkcnt:
398 erofs_err(sb, "bogus CBLKCNT @ lcn %lu of nid %llu", lcn, vi->nid);
399 DBG_BUGON(1);
400 return -EFSCORRUPTED;
401}
402
403static int z_erofs_get_extent_decompressedlen(struct z_erofs_maprecorder *m)
404{
405 struct inode *inode = m->inode;
406 struct erofs_inode *vi = EROFS_I(inode);
407 struct erofs_map_blocks *map = m->map;
408 unsigned int lclusterbits = vi->z_logical_clusterbits;
409 u64 lcn = m->lcn, headlcn = map->m_la >> lclusterbits;
410 int err;
411
412 do {
413 /* handle the last EOF pcluster (no next HEAD lcluster) */
414 if ((lcn << lclusterbits) >= inode->i_size) {
415 map->m_llen = inode->i_size - map->m_la;
416 return 0;
417 }
418
419 err = z_erofs_load_lcluster_from_disk(m, lcn, true);
420 if (err)
421 return err;
422
423 if (m->type == Z_EROFS_LCLUSTER_TYPE_NONHEAD) {
424 DBG_BUGON(!m->delta[1] &&
425 m->clusterofs != 1 << lclusterbits);
426 } else if (m->type == Z_EROFS_LCLUSTER_TYPE_PLAIN ||
427 m->type == Z_EROFS_LCLUSTER_TYPE_HEAD1 ||
428 m->type == Z_EROFS_LCLUSTER_TYPE_HEAD2) {
429 /* go on until the next HEAD lcluster */
430 if (lcn != headlcn)
431 break;
432 m->delta[1] = 1;
433 } else {
434 erofs_err(inode->i_sb, "unknown type %u @ lcn %llu of nid %llu",
435 m->type, lcn, vi->nid);
436 DBG_BUGON(1);
437 return -EOPNOTSUPP;
438 }
439 lcn += m->delta[1];
440 } while (m->delta[1]);
441
442 map->m_llen = (lcn << lclusterbits) + m->clusterofs - map->m_la;
443 return 0;
444}
445
446static int z_erofs_do_map_blocks(struct inode *inode,
447 struct erofs_map_blocks *map, int flags)
448{
449 struct erofs_inode *const vi = EROFS_I(inode);
450 bool ztailpacking = vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER;
451 bool fragment = vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
452 struct z_erofs_maprecorder m = {
453 .inode = inode,
454 .map = map,
455 };
456 int err = 0;
457 unsigned int lclusterbits, endoff, afmt;
458 unsigned long initial_lcn;
459 unsigned long long ofs, end;
460
461 lclusterbits = vi->z_logical_clusterbits;
462 ofs = flags & EROFS_GET_BLOCKS_FINDTAIL ? inode->i_size - 1 : map->m_la;
463 initial_lcn = ofs >> lclusterbits;
464 endoff = ofs & ((1 << lclusterbits) - 1);
465
466 err = z_erofs_load_lcluster_from_disk(&m, initial_lcn, false);
467 if (err)
468 goto unmap_out;
469
470 if (ztailpacking && (flags & EROFS_GET_BLOCKS_FINDTAIL))
471 vi->z_idataoff = m.nextpackoff;
472
473 map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_ENCODED;
474 end = (m.lcn + 1ULL) << lclusterbits;
475
476 switch (m.type) {
477 case Z_EROFS_LCLUSTER_TYPE_PLAIN:
478 case Z_EROFS_LCLUSTER_TYPE_HEAD1:
479 case Z_EROFS_LCLUSTER_TYPE_HEAD2:
480 if (endoff >= m.clusterofs) {
481 m.headtype = m.type;
482 map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
483 /*
484 * For ztailpacking files, in order to inline data more
485 * effectively, special EOF lclusters are now supported
486 * which can have three parts at most.
487 */
488 if (ztailpacking && end > inode->i_size)
489 end = inode->i_size;
490 break;
491 }
492 /* m.lcn should be >= 1 if endoff < m.clusterofs */
493 if (!m.lcn) {
494 erofs_err(inode->i_sb,
495 "invalid logical cluster 0 at nid %llu",
496 vi->nid);
497 err = -EFSCORRUPTED;
498 goto unmap_out;
499 }
500 end = (m.lcn << lclusterbits) | m.clusterofs;
501 map->m_flags |= EROFS_MAP_FULL_MAPPED;
502 m.delta[0] = 1;
503 fallthrough;
504 case Z_EROFS_LCLUSTER_TYPE_NONHEAD:
505 /* get the corresponding first chunk */
506 err = z_erofs_extent_lookback(&m, m.delta[0]);
507 if (err)
508 goto unmap_out;
509 break;
510 default:
511 erofs_err(inode->i_sb,
512 "unknown type %u @ offset %llu of nid %llu",
513 m.type, ofs, vi->nid);
514 err = -EOPNOTSUPP;
515 goto unmap_out;
516 }
517 if (m.partialref)
518 map->m_flags |= EROFS_MAP_PARTIAL_REF;
519 map->m_llen = end - map->m_la;
520
521 if (flags & EROFS_GET_BLOCKS_FINDTAIL) {
522 vi->z_tailextent_headlcn = m.lcn;
523 /* for non-compact indexes, fragmentoff is 64 bits */
524 if (fragment && vi->datalayout == EROFS_INODE_COMPRESSED_FULL)
525 vi->z_fragmentoff |= (u64)m.pblk << 32;
526 }
527 if (ztailpacking && m.lcn == vi->z_tailextent_headlcn) {
528 map->m_flags |= EROFS_MAP_META;
529 map->m_pa = vi->z_idataoff;
530 map->m_plen = vi->z_idata_size;
531 } else if (fragment && m.lcn == vi->z_tailextent_headlcn) {
532 map->m_flags |= EROFS_MAP_FRAGMENT;
533 } else {
534 map->m_pa = erofs_pos(inode->i_sb, m.pblk);
535 err = z_erofs_get_extent_compressedlen(&m, initial_lcn);
536 if (err)
537 goto unmap_out;
538 }
539
540 if (m.headtype == Z_EROFS_LCLUSTER_TYPE_PLAIN) {
541 if (map->m_llen > map->m_plen) {
542 DBG_BUGON(1);
543 err = -EFSCORRUPTED;
544 goto unmap_out;
545 }
546 afmt = vi->z_advise & Z_EROFS_ADVISE_INTERLACED_PCLUSTER ?
547 Z_EROFS_COMPRESSION_INTERLACED :
548 Z_EROFS_COMPRESSION_SHIFTED;
549 } else {
550 afmt = m.headtype == Z_EROFS_LCLUSTER_TYPE_HEAD2 ?
551 vi->z_algorithmtype[1] : vi->z_algorithmtype[0];
552 if (!(EROFS_I_SB(inode)->available_compr_algs & (1 << afmt))) {
553 erofs_err(inode->i_sb, "inconsistent algorithmtype %u for nid %llu",
554 afmt, vi->nid);
555 err = -EFSCORRUPTED;
556 goto unmap_out;
557 }
558 }
559 map->m_algorithmformat = afmt;
560
561 if ((flags & EROFS_GET_BLOCKS_FIEMAP) ||
562 ((flags & EROFS_GET_BLOCKS_READMORE) &&
563 (map->m_algorithmformat == Z_EROFS_COMPRESSION_LZMA ||
564 map->m_algorithmformat == Z_EROFS_COMPRESSION_DEFLATE) &&
565 map->m_llen >= i_blocksize(inode))) {
566 err = z_erofs_get_extent_decompressedlen(&m);
567 if (!err)
568 map->m_flags |= EROFS_MAP_FULL_MAPPED;
569 }
570
571unmap_out:
572 erofs_unmap_metabuf(&m.map->buf);
573 return err;
574}
575
576static int z_erofs_fill_inode_lazy(struct inode *inode)
577{
578 struct erofs_inode *const vi = EROFS_I(inode);
579 struct super_block *const sb = inode->i_sb;
580 int err, headnr;
581 erofs_off_t pos;
582 struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
583 void *kaddr;
584 struct z_erofs_map_header *h;
585
586 if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags)) {
587 /*
588 * paired with smp_mb() at the end of the function to ensure
589 * fields will only be observed after the bit is set.
590 */
591 smp_mb();
592 return 0;
593 }
594
595 if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE))
596 return -ERESTARTSYS;
597
598 err = 0;
599 if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
600 goto out_unlock;
601
602 pos = ALIGN(erofs_iloc(inode) + vi->inode_isize + vi->xattr_isize, 8);
603 kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(sb, pos), EROFS_KMAP);
604 if (IS_ERR(kaddr)) {
605 err = PTR_ERR(kaddr);
606 goto out_unlock;
607 }
608
609 h = kaddr + erofs_blkoff(sb, pos);
610 /*
611 * if the highest bit of the 8-byte map header is set, the whole file
612 * is stored in the packed inode. The rest bits keeps z_fragmentoff.
613 */
614 if (h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT) {
615 vi->z_advise = Z_EROFS_ADVISE_FRAGMENT_PCLUSTER;
616 vi->z_fragmentoff = le64_to_cpu(*(__le64 *)h) ^ (1ULL << 63);
617 vi->z_tailextent_headlcn = 0;
618 goto done;
619 }
620 vi->z_advise = le16_to_cpu(h->h_advise);
621 vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
622 vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
623
624 headnr = 0;
625 if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX ||
626 vi->z_algorithmtype[++headnr] >= Z_EROFS_COMPRESSION_MAX) {
627 erofs_err(sb, "unknown HEAD%u format %u for nid %llu, please upgrade kernel",
628 headnr + 1, vi->z_algorithmtype[headnr], vi->nid);
629 err = -EOPNOTSUPP;
630 goto out_put_metabuf;
631 }
632
633 vi->z_logical_clusterbits = sb->s_blocksize_bits + (h->h_clusterbits & 7);
634 if (!erofs_sb_has_big_pcluster(EROFS_SB(sb)) &&
635 vi->z_advise & (Z_EROFS_ADVISE_BIG_PCLUSTER_1 |
636 Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
637 erofs_err(sb, "per-inode big pcluster without sb feature for nid %llu",
638 vi->nid);
639 err = -EFSCORRUPTED;
640 goto out_put_metabuf;
641 }
642 if (vi->datalayout == EROFS_INODE_COMPRESSED_COMPACT &&
643 !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_1) ^
644 !(vi->z_advise & Z_EROFS_ADVISE_BIG_PCLUSTER_2)) {
645 erofs_err(sb, "big pcluster head1/2 of compact indexes should be consistent for nid %llu",
646 vi->nid);
647 err = -EFSCORRUPTED;
648 goto out_put_metabuf;
649 }
650
651 if (vi->z_advise & Z_EROFS_ADVISE_INLINE_PCLUSTER) {
652 struct erofs_map_blocks map = {
653 .buf = __EROFS_BUF_INITIALIZER
654 };
655
656 vi->z_idata_size = le16_to_cpu(h->h_idata_size);
657 err = z_erofs_do_map_blocks(inode, &map,
658 EROFS_GET_BLOCKS_FINDTAIL);
659 erofs_put_metabuf(&map.buf);
660
661 if (!map.m_plen ||
662 erofs_blkoff(sb, map.m_pa) + map.m_plen > sb->s_blocksize) {
663 erofs_err(sb, "invalid tail-packing pclustersize %llu",
664 map.m_plen);
665 err = -EFSCORRUPTED;
666 }
667 if (err < 0)
668 goto out_put_metabuf;
669 }
670
671 if (vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER &&
672 !(h->h_clusterbits >> Z_EROFS_FRAGMENT_INODE_BIT)) {
673 struct erofs_map_blocks map = {
674 .buf = __EROFS_BUF_INITIALIZER
675 };
676
677 vi->z_fragmentoff = le32_to_cpu(h->h_fragmentoff);
678 err = z_erofs_do_map_blocks(inode, &map,
679 EROFS_GET_BLOCKS_FINDTAIL);
680 erofs_put_metabuf(&map.buf);
681 if (err < 0)
682 goto out_put_metabuf;
683 }
684done:
685 /* paired with smp_mb() at the beginning of the function */
686 smp_mb();
687 set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
688out_put_metabuf:
689 erofs_put_metabuf(&buf);
690out_unlock:
691 clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
692 return err;
693}
694
695int z_erofs_map_blocks_iter(struct inode *inode, struct erofs_map_blocks *map,
696 int flags)
697{
698 struct erofs_inode *const vi = EROFS_I(inode);
699 int err = 0;
700
701 trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
702
703 /* when trying to read beyond EOF, leave it unmapped */
704 if (map->m_la >= inode->i_size) {
705 map->m_llen = map->m_la + 1 - inode->i_size;
706 map->m_la = inode->i_size;
707 map->m_flags = 0;
708 goto out;
709 }
710
711 err = z_erofs_fill_inode_lazy(inode);
712 if (err)
713 goto out;
714
715 if ((vi->z_advise & Z_EROFS_ADVISE_FRAGMENT_PCLUSTER) &&
716 !vi->z_tailextent_headlcn) {
717 map->m_la = 0;
718 map->m_llen = inode->i_size;
719 map->m_flags = EROFS_MAP_MAPPED | EROFS_MAP_FULL_MAPPED |
720 EROFS_MAP_FRAGMENT;
721 goto out;
722 }
723
724 err = z_erofs_do_map_blocks(inode, map, flags);
725out:
726 trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
727 return err;
728}
729
730static int z_erofs_iomap_begin_report(struct inode *inode, loff_t offset,
731 loff_t length, unsigned int flags,
732 struct iomap *iomap, struct iomap *srcmap)
733{
734 int ret;
735 struct erofs_map_blocks map = { .m_la = offset };
736
737 ret = z_erofs_map_blocks_iter(inode, &map, EROFS_GET_BLOCKS_FIEMAP);
738 erofs_put_metabuf(&map.buf);
739 if (ret < 0)
740 return ret;
741
742 iomap->bdev = inode->i_sb->s_bdev;
743 iomap->offset = map.m_la;
744 iomap->length = map.m_llen;
745 if (map.m_flags & EROFS_MAP_MAPPED) {
746 iomap->type = IOMAP_MAPPED;
747 iomap->addr = map.m_flags & EROFS_MAP_FRAGMENT ?
748 IOMAP_NULL_ADDR : map.m_pa;
749 } else {
750 iomap->type = IOMAP_HOLE;
751 iomap->addr = IOMAP_NULL_ADDR;
752 /*
753 * No strict rule on how to describe extents for post EOF, yet
754 * we need to do like below. Otherwise, iomap itself will get
755 * into an endless loop on post EOF.
756 *
757 * Calculate the effective offset by subtracting extent start
758 * (map.m_la) from the requested offset, and add it to length.
759 * (NB: offset >= map.m_la always)
760 */
761 if (iomap->offset >= inode->i_size)
762 iomap->length = length + offset - map.m_la;
763 }
764 iomap->flags = 0;
765 return 0;
766}
767
768const struct iomap_ops z_erofs_iomap_report_ops = {
769 .iomap_begin = z_erofs_iomap_begin_report,
770};
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
10static int z_erofs_do_map_blocks(struct inode *inode,
11 struct erofs_map_blocks *map,
12 int flags);
13
14int 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
32static 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 }
141done:
142 /* paired with smp_mb() at the beginning of the function */
143 smp_mb();
144 set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
145out_put_metabuf:
146 erofs_put_metabuf(&buf);
147out_unlock:
148 clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
149 return err;
150}
151
152struct 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
167static 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
224static 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
235static 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
260static 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
369static 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;
416out:
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
425static 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
439static 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
486static 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 }
550out:
551 map->m_plen = (u64)m->compressedblks << LOG_BLOCK_SIZE;
552 return 0;
553err_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
561static 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
604static 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
727unmap_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
735int 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);
765out:
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
773static 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
811const struct iomap_ops z_erofs_iomap_report_ops = {
812 .iomap_begin = z_erofs_iomap_begin_report,
813};