xref: /openbmc/linux/fs/erofs/zmap.c (revision 3a83e4e6)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2018-2019 HUAWEI, Inc.
4  *             https://www.huawei.com/
5  * Created by Gao Xiang <gaoxiang25@huawei.com>
6  */
7 #include "internal.h"
8 #include <asm/unaligned.h>
9 #include <trace/events/erofs.h>
10 
11 int z_erofs_fill_inode(struct inode *inode)
12 {
13 	struct erofs_inode *const vi = EROFS_I(inode);
14 
15 	if (vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY) {
16 		vi->z_advise = 0;
17 		vi->z_algorithmtype[0] = 0;
18 		vi->z_algorithmtype[1] = 0;
19 		vi->z_logical_clusterbits = LOG_BLOCK_SIZE;
20 		vi->z_physical_clusterbits[0] = vi->z_logical_clusterbits;
21 		vi->z_physical_clusterbits[1] = vi->z_logical_clusterbits;
22 		set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
23 	}
24 
25 	inode->i_mapping->a_ops = &z_erofs_aops;
26 	return 0;
27 }
28 
29 static int z_erofs_fill_inode_lazy(struct inode *inode)
30 {
31 	struct erofs_inode *const vi = EROFS_I(inode);
32 	struct super_block *const sb = inode->i_sb;
33 	int err;
34 	erofs_off_t pos;
35 	struct page *page;
36 	void *kaddr;
37 	struct z_erofs_map_header *h;
38 
39 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
40 		return 0;
41 
42 	if (wait_on_bit_lock(&vi->flags, EROFS_I_BL_Z_BIT, TASK_KILLABLE))
43 		return -ERESTARTSYS;
44 
45 	err = 0;
46 	if (test_bit(EROFS_I_Z_INITED_BIT, &vi->flags))
47 		goto out_unlock;
48 
49 	DBG_BUGON(vi->datalayout == EROFS_INODE_FLAT_COMPRESSION_LEGACY);
50 
51 	pos = ALIGN(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
52 		    vi->xattr_isize, 8);
53 	page = erofs_get_meta_page(sb, erofs_blknr(pos));
54 	if (IS_ERR(page)) {
55 		err = PTR_ERR(page);
56 		goto out_unlock;
57 	}
58 
59 	kaddr = kmap_atomic(page);
60 
61 	h = kaddr + erofs_blkoff(pos);
62 	vi->z_advise = le16_to_cpu(h->h_advise);
63 	vi->z_algorithmtype[0] = h->h_algorithmtype & 15;
64 	vi->z_algorithmtype[1] = h->h_algorithmtype >> 4;
65 
66 	if (vi->z_algorithmtype[0] >= Z_EROFS_COMPRESSION_MAX) {
67 		erofs_err(sb, "unknown compression format %u for nid %llu, please upgrade kernel",
68 			  vi->z_algorithmtype[0], vi->nid);
69 		err = -EOPNOTSUPP;
70 		goto unmap_done;
71 	}
72 
73 	vi->z_logical_clusterbits = LOG_BLOCK_SIZE + (h->h_clusterbits & 7);
74 	vi->z_physical_clusterbits[0] = vi->z_logical_clusterbits +
75 					((h->h_clusterbits >> 3) & 3);
76 
77 	if (vi->z_physical_clusterbits[0] != LOG_BLOCK_SIZE) {
78 		erofs_err(sb, "unsupported physical clusterbits %u for nid %llu, please upgrade kernel",
79 			  vi->z_physical_clusterbits[0], vi->nid);
80 		err = -EOPNOTSUPP;
81 		goto unmap_done;
82 	}
83 
84 	vi->z_physical_clusterbits[1] = vi->z_logical_clusterbits +
85 					((h->h_clusterbits >> 5) & 7);
86 	set_bit(EROFS_I_Z_INITED_BIT, &vi->flags);
87 unmap_done:
88 	kunmap_atomic(kaddr);
89 	unlock_page(page);
90 	put_page(page);
91 out_unlock:
92 	clear_and_wake_up_bit(EROFS_I_BL_Z_BIT, &vi->flags);
93 	return err;
94 }
95 
96 struct z_erofs_maprecorder {
97 	struct inode *inode;
98 	struct erofs_map_blocks *map;
99 	void *kaddr;
100 
101 	unsigned long lcn;
102 	/* compression extent information gathered */
103 	u8  type;
104 	u16 clusterofs;
105 	u16 delta[2];
106 	erofs_blk_t pblk;
107 };
108 
109 static int z_erofs_reload_indexes(struct z_erofs_maprecorder *m,
110 				  erofs_blk_t eblk)
111 {
112 	struct super_block *const sb = m->inode->i_sb;
113 	struct erofs_map_blocks *const map = m->map;
114 	struct page *mpage = map->mpage;
115 
116 	if (mpage) {
117 		if (mpage->index == eblk) {
118 			if (!m->kaddr)
119 				m->kaddr = kmap_atomic(mpage);
120 			return 0;
121 		}
122 
123 		if (m->kaddr) {
124 			kunmap_atomic(m->kaddr);
125 			m->kaddr = NULL;
126 		}
127 		put_page(mpage);
128 	}
129 
130 	mpage = erofs_get_meta_page(sb, eblk);
131 	if (IS_ERR(mpage)) {
132 		map->mpage = NULL;
133 		return PTR_ERR(mpage);
134 	}
135 	m->kaddr = kmap_atomic(mpage);
136 	unlock_page(mpage);
137 	map->mpage = mpage;
138 	return 0;
139 }
140 
141 static int legacy_load_cluster_from_disk(struct z_erofs_maprecorder *m,
142 					 unsigned long lcn)
143 {
144 	struct inode *const inode = m->inode;
145 	struct erofs_inode *const vi = EROFS_I(inode);
146 	const erofs_off_t ibase = iloc(EROFS_I_SB(inode), vi->nid);
147 	const erofs_off_t pos =
148 		Z_EROFS_VLE_LEGACY_INDEX_ALIGN(ibase + vi->inode_isize +
149 					       vi->xattr_isize) +
150 		lcn * sizeof(struct z_erofs_vle_decompressed_index);
151 	struct z_erofs_vle_decompressed_index *di;
152 	unsigned int advise, type;
153 	int err;
154 
155 	err = z_erofs_reload_indexes(m, erofs_blknr(pos));
156 	if (err)
157 		return err;
158 
159 	m->lcn = lcn;
160 	di = m->kaddr + erofs_blkoff(pos);
161 
162 	advise = le16_to_cpu(di->di_advise);
163 	type = (advise >> Z_EROFS_VLE_DI_CLUSTER_TYPE_BIT) &
164 		((1 << Z_EROFS_VLE_DI_CLUSTER_TYPE_BITS) - 1);
165 	switch (type) {
166 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
167 		m->clusterofs = 1 << vi->z_logical_clusterbits;
168 		m->delta[0] = le16_to_cpu(di->di_u.delta[0]);
169 		m->delta[1] = le16_to_cpu(di->di_u.delta[1]);
170 		break;
171 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
172 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD:
173 		m->clusterofs = le16_to_cpu(di->di_clusterofs);
174 		m->pblk = le32_to_cpu(di->di_u.blkaddr);
175 		break;
176 	default:
177 		DBG_BUGON(1);
178 		return -EOPNOTSUPP;
179 	}
180 	m->type = type;
181 	return 0;
182 }
183 
184 static unsigned int decode_compactedbits(unsigned int lobits,
185 					 unsigned int lomask,
186 					 u8 *in, unsigned int pos, u8 *type)
187 {
188 	const unsigned int v = get_unaligned_le32(in + pos / 8) >> (pos & 7);
189 	const unsigned int lo = v & lomask;
190 
191 	*type = (v >> lobits) & 3;
192 	return lo;
193 }
194 
195 static int unpack_compacted_index(struct z_erofs_maprecorder *m,
196 				  unsigned int amortizedshift,
197 				  unsigned int eofs)
198 {
199 	struct erofs_inode *const vi = EROFS_I(m->inode);
200 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
201 	const unsigned int lomask = (1 << lclusterbits) - 1;
202 	unsigned int vcnt, base, lo, encodebits, nblk;
203 	int i;
204 	u8 *in, type;
205 
206 	if (1 << amortizedshift == 4)
207 		vcnt = 2;
208 	else if (1 << amortizedshift == 2 && lclusterbits == 12)
209 		vcnt = 16;
210 	else
211 		return -EOPNOTSUPP;
212 
213 	encodebits = ((vcnt << amortizedshift) - sizeof(__le32)) * 8 / vcnt;
214 	base = round_down(eofs, vcnt << amortizedshift);
215 	in = m->kaddr + base;
216 
217 	i = (eofs - base) >> amortizedshift;
218 
219 	lo = decode_compactedbits(lclusterbits, lomask,
220 				  in, encodebits * i, &type);
221 	m->type = type;
222 	if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD) {
223 		m->clusterofs = 1 << lclusterbits;
224 		if (i + 1 != vcnt) {
225 			m->delta[0] = lo;
226 			return 0;
227 		}
228 		/*
229 		 * since the last lcluster in the pack is special,
230 		 * of which lo saves delta[1] rather than delta[0].
231 		 * Hence, get delta[0] by the previous lcluster indirectly.
232 		 */
233 		lo = decode_compactedbits(lclusterbits, lomask,
234 					  in, encodebits * (i - 1), &type);
235 		if (type != Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
236 			lo = 0;
237 		m->delta[0] = lo + 1;
238 		return 0;
239 	}
240 	m->clusterofs = lo;
241 	m->delta[0] = 0;
242 	/* figout out blkaddr (pblk) for HEAD lclusters */
243 	nblk = 1;
244 	while (i > 0) {
245 		--i;
246 		lo = decode_compactedbits(lclusterbits, lomask,
247 					  in, encodebits * i, &type);
248 		if (type == Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD)
249 			i -= lo;
250 
251 		if (i >= 0)
252 			++nblk;
253 	}
254 	in += (vcnt << amortizedshift) - sizeof(__le32);
255 	m->pblk = le32_to_cpu(*(__le32 *)in) + nblk;
256 	return 0;
257 }
258 
259 static int compacted_load_cluster_from_disk(struct z_erofs_maprecorder *m,
260 					    unsigned long lcn)
261 {
262 	struct inode *const inode = m->inode;
263 	struct erofs_inode *const vi = EROFS_I(inode);
264 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
265 	const erofs_off_t ebase = ALIGN(iloc(EROFS_I_SB(inode), vi->nid) +
266 					vi->inode_isize + vi->xattr_isize, 8) +
267 		sizeof(struct z_erofs_map_header);
268 	const unsigned int totalidx = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
269 	unsigned int compacted_4b_initial, compacted_2b;
270 	unsigned int amortizedshift;
271 	erofs_off_t pos;
272 	int err;
273 
274 	if (lclusterbits != 12)
275 		return -EOPNOTSUPP;
276 
277 	if (lcn >= totalidx)
278 		return -EINVAL;
279 
280 	m->lcn = lcn;
281 	/* used to align to 32-byte (compacted_2b) alignment */
282 	compacted_4b_initial = (32 - ebase % 32) / 4;
283 	if (compacted_4b_initial == 32 / 4)
284 		compacted_4b_initial = 0;
285 
286 	if (vi->z_advise & Z_EROFS_ADVISE_COMPACTED_2B)
287 		compacted_2b = rounddown(totalidx - compacted_4b_initial, 16);
288 	else
289 		compacted_2b = 0;
290 
291 	pos = ebase;
292 	if (lcn < compacted_4b_initial) {
293 		amortizedshift = 2;
294 		goto out;
295 	}
296 	pos += compacted_4b_initial * 4;
297 	lcn -= compacted_4b_initial;
298 
299 	if (lcn < compacted_2b) {
300 		amortizedshift = 1;
301 		goto out;
302 	}
303 	pos += compacted_2b * 2;
304 	lcn -= compacted_2b;
305 	amortizedshift = 2;
306 out:
307 	pos += lcn * (1 << amortizedshift);
308 	err = z_erofs_reload_indexes(m, erofs_blknr(pos));
309 	if (err)
310 		return err;
311 	return unpack_compacted_index(m, amortizedshift, erofs_blkoff(pos));
312 }
313 
314 static int z_erofs_load_cluster_from_disk(struct z_erofs_maprecorder *m,
315 					  unsigned int lcn)
316 {
317 	const unsigned int datamode = EROFS_I(m->inode)->datalayout;
318 
319 	if (datamode == EROFS_INODE_FLAT_COMPRESSION_LEGACY)
320 		return legacy_load_cluster_from_disk(m, lcn);
321 
322 	if (datamode == EROFS_INODE_FLAT_COMPRESSION)
323 		return compacted_load_cluster_from_disk(m, lcn);
324 
325 	return -EINVAL;
326 }
327 
328 static int z_erofs_extent_lookback(struct z_erofs_maprecorder *m,
329 				   unsigned int lookback_distance)
330 {
331 	struct erofs_inode *const vi = EROFS_I(m->inode);
332 	struct erofs_map_blocks *const map = m->map;
333 	const unsigned int lclusterbits = vi->z_logical_clusterbits;
334 	unsigned long lcn = m->lcn;
335 	int err;
336 
337 	if (lcn < lookback_distance) {
338 		erofs_err(m->inode->i_sb,
339 			  "bogus lookback distance @ nid %llu", vi->nid);
340 		DBG_BUGON(1);
341 		return -EFSCORRUPTED;
342 	}
343 
344 	/* load extent head logical cluster if needed */
345 	lcn -= lookback_distance;
346 	err = z_erofs_load_cluster_from_disk(m, lcn);
347 	if (err)
348 		return err;
349 
350 	switch (m->type) {
351 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
352 		if (!m->delta[0]) {
353 			erofs_err(m->inode->i_sb,
354 				  "invalid lookback distance 0 @ nid %llu",
355 				  vi->nid);
356 			DBG_BUGON(1);
357 			return -EFSCORRUPTED;
358 		}
359 		return z_erofs_extent_lookback(m, m->delta[0]);
360 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
361 		map->m_flags &= ~EROFS_MAP_ZIPPED;
362 		fallthrough;
363 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD:
364 		map->m_la = (lcn << lclusterbits) | m->clusterofs;
365 		break;
366 	default:
367 		erofs_err(m->inode->i_sb,
368 			  "unknown type %u @ lcn %lu of nid %llu",
369 			  m->type, lcn, vi->nid);
370 		DBG_BUGON(1);
371 		return -EOPNOTSUPP;
372 	}
373 	return 0;
374 }
375 
376 int z_erofs_map_blocks_iter(struct inode *inode,
377 			    struct erofs_map_blocks *map,
378 			    int flags)
379 {
380 	struct erofs_inode *const vi = EROFS_I(inode);
381 	struct z_erofs_maprecorder m = {
382 		.inode = inode,
383 		.map = map,
384 	};
385 	int err = 0;
386 	unsigned int lclusterbits, endoff;
387 	unsigned long long ofs, end;
388 
389 	trace_z_erofs_map_blocks_iter_enter(inode, map, flags);
390 
391 	/* when trying to read beyond EOF, leave it unmapped */
392 	if (map->m_la >= inode->i_size) {
393 		map->m_llen = map->m_la + 1 - inode->i_size;
394 		map->m_la = inode->i_size;
395 		map->m_flags = 0;
396 		goto out;
397 	}
398 
399 	err = z_erofs_fill_inode_lazy(inode);
400 	if (err)
401 		goto out;
402 
403 	lclusterbits = vi->z_logical_clusterbits;
404 	ofs = map->m_la;
405 	m.lcn = ofs >> lclusterbits;
406 	endoff = ofs & ((1 << lclusterbits) - 1);
407 
408 	err = z_erofs_load_cluster_from_disk(&m, m.lcn);
409 	if (err)
410 		goto unmap_out;
411 
412 	map->m_flags = EROFS_MAP_ZIPPED;	/* by default, compressed */
413 	end = (m.lcn + 1ULL) << lclusterbits;
414 
415 	switch (m.type) {
416 	case Z_EROFS_VLE_CLUSTER_TYPE_PLAIN:
417 		if (endoff >= m.clusterofs)
418 			map->m_flags &= ~EROFS_MAP_ZIPPED;
419 		fallthrough;
420 	case Z_EROFS_VLE_CLUSTER_TYPE_HEAD:
421 		if (endoff >= m.clusterofs) {
422 			map->m_la = (m.lcn << lclusterbits) | m.clusterofs;
423 			break;
424 		}
425 		/* m.lcn should be >= 1 if endoff < m.clusterofs */
426 		if (!m.lcn) {
427 			erofs_err(inode->i_sb,
428 				  "invalid logical cluster 0 at nid %llu",
429 				  vi->nid);
430 			err = -EFSCORRUPTED;
431 			goto unmap_out;
432 		}
433 		end = (m.lcn << lclusterbits) | m.clusterofs;
434 		map->m_flags |= EROFS_MAP_FULL_MAPPED;
435 		m.delta[0] = 1;
436 		fallthrough;
437 	case Z_EROFS_VLE_CLUSTER_TYPE_NONHEAD:
438 		/* get the correspoinding first chunk */
439 		err = z_erofs_extent_lookback(&m, m.delta[0]);
440 		if (err)
441 			goto unmap_out;
442 		break;
443 	default:
444 		erofs_err(inode->i_sb,
445 			  "unknown type %u @ offset %llu of nid %llu",
446 			  m.type, ofs, vi->nid);
447 		err = -EOPNOTSUPP;
448 		goto unmap_out;
449 	}
450 
451 	map->m_llen = end - map->m_la;
452 	map->m_plen = 1 << lclusterbits;
453 	map->m_pa = blknr_to_addr(m.pblk);
454 	map->m_flags |= EROFS_MAP_MAPPED;
455 
456 unmap_out:
457 	if (m.kaddr)
458 		kunmap_atomic(m.kaddr);
459 
460 out:
461 	erofs_dbg("%s, m_la %llu m_pa %llu m_llen %llu m_plen %llu m_flags 0%o",
462 		  __func__, map->m_la, map->m_pa,
463 		  map->m_llen, map->m_plen, map->m_flags);
464 
465 	trace_z_erofs_map_blocks_iter_exit(inode, map, flags, err);
466 
467 	/* aggressively BUG_ON iff CONFIG_EROFS_FS_DEBUG is on */
468 	DBG_BUGON(err < 0 && err != -ENOMEM);
469 	return err;
470 }
471 
472