xref: /linux/fs/erofs/data.c (revision 55d0969c451159cff86949b38c39171cab962069) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017-2018 HUAWEI, Inc.
4  *             https://www.huawei.com/
5  * Copyright (C) 2021, Alibaba Cloud
6  */
7 #include "internal.h"
8 #include <linux/sched/mm.h>
9 #include <trace/events/erofs.h>
10 
11 void erofs_unmap_metabuf(struct erofs_buf *buf)
12 {
13 	if (buf->kmap_type == EROFS_KMAP)
14 		kunmap_local(buf->base);
15 	buf->base = NULL;
16 	buf->kmap_type = EROFS_NO_KMAP;
17 }
18 
19 void erofs_put_metabuf(struct erofs_buf *buf)
20 {
21 	if (!buf->page)
22 		return;
23 	erofs_unmap_metabuf(buf);
24 	folio_put(page_folio(buf->page));
25 	buf->page = NULL;
26 }
27 
28 void *erofs_bread(struct erofs_buf *buf, erofs_off_t offset,
29 		  enum erofs_kmap_type type)
30 {
31 	pgoff_t index = offset >> PAGE_SHIFT;
32 	struct folio *folio = NULL;
33 
34 	if (buf->page) {
35 		folio = page_folio(buf->page);
36 		if (folio_file_page(folio, index) != buf->page)
37 			erofs_unmap_metabuf(buf);
38 	}
39 	if (!folio || !folio_contains(folio, index)) {
40 		erofs_put_metabuf(buf);
41 		folio = read_mapping_folio(buf->mapping, index, NULL);
42 		if (IS_ERR(folio))
43 			return folio;
44 	}
45 	buf->page = folio_file_page(folio, index);
46 
47 	if (buf->kmap_type == EROFS_NO_KMAP) {
48 		if (type == EROFS_KMAP)
49 			buf->base = kmap_local_page(buf->page);
50 		buf->kmap_type = type;
51 	} else if (buf->kmap_type != type) {
52 		DBG_BUGON(1);
53 		return ERR_PTR(-EFAULT);
54 	}
55 	if (type == EROFS_NO_KMAP)
56 		return NULL;
57 	return buf->base + (offset & ~PAGE_MASK);
58 }
59 
60 void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
61 {
62 	struct erofs_sb_info *sbi = EROFS_SB(sb);
63 
64 	if (erofs_is_fileio_mode(sbi))
65 		buf->mapping = file_inode(sbi->fdev)->i_mapping;
66 	else if (erofs_is_fscache_mode(sb))
67 		buf->mapping = sbi->s_fscache->inode->i_mapping;
68 	else
69 		buf->mapping = sb->s_bdev->bd_mapping;
70 }
71 
72 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
73 			 erofs_off_t offset, enum erofs_kmap_type type)
74 {
75 	erofs_init_metabuf(buf, sb);
76 	return erofs_bread(buf, offset, type);
77 }
78 
79 static int erofs_map_blocks_flatmode(struct inode *inode,
80 				     struct erofs_map_blocks *map)
81 {
82 	struct erofs_inode *vi = EROFS_I(inode);
83 	struct super_block *sb = inode->i_sb;
84 	bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
85 	erofs_blk_t lastblk = erofs_iblks(inode) - tailendpacking;
86 
87 	map->m_flags = EROFS_MAP_MAPPED;	/* no hole in flat inodes */
88 	if (map->m_la < erofs_pos(sb, lastblk)) {
89 		map->m_pa = erofs_pos(sb, vi->raw_blkaddr) + map->m_la;
90 		map->m_plen = erofs_pos(sb, lastblk) - map->m_la;
91 	} else {
92 		DBG_BUGON(!tailendpacking);
93 		map->m_pa = erofs_iloc(inode) + vi->inode_isize +
94 			vi->xattr_isize + erofs_blkoff(sb, map->m_la);
95 		map->m_plen = inode->i_size - map->m_la;
96 
97 		/* inline data should be located in the same meta block */
98 		if (erofs_blkoff(sb, map->m_pa) + map->m_plen > sb->s_blocksize) {
99 			erofs_err(sb, "inline data across blocks @ nid %llu", vi->nid);
100 			DBG_BUGON(1);
101 			return -EFSCORRUPTED;
102 		}
103 		map->m_flags |= EROFS_MAP_META;
104 	}
105 	return 0;
106 }
107 
108 int erofs_map_blocks(struct inode *inode, struct erofs_map_blocks *map)
109 {
110 	struct super_block *sb = inode->i_sb;
111 	struct erofs_inode *vi = EROFS_I(inode);
112 	struct erofs_inode_chunk_index *idx;
113 	struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
114 	u64 chunknr;
115 	unsigned int unit;
116 	erofs_off_t pos;
117 	void *kaddr;
118 	int err = 0;
119 
120 	trace_erofs_map_blocks_enter(inode, map, 0);
121 	map->m_deviceid = 0;
122 	if (map->m_la >= inode->i_size) {
123 		/* leave out-of-bound access unmapped */
124 		map->m_flags = 0;
125 		map->m_plen = map->m_llen;
126 		goto out;
127 	}
128 
129 	if (vi->datalayout != EROFS_INODE_CHUNK_BASED) {
130 		err = erofs_map_blocks_flatmode(inode, map);
131 		goto out;
132 	}
133 
134 	if (vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)
135 		unit = sizeof(*idx);			/* chunk index */
136 	else
137 		unit = EROFS_BLOCK_MAP_ENTRY_SIZE;	/* block map */
138 
139 	chunknr = map->m_la >> vi->chunkbits;
140 	pos = ALIGN(erofs_iloc(inode) + vi->inode_isize +
141 		    vi->xattr_isize, unit) + unit * chunknr;
142 
143 	kaddr = erofs_read_metabuf(&buf, sb, pos, EROFS_KMAP);
144 	if (IS_ERR(kaddr)) {
145 		err = PTR_ERR(kaddr);
146 		goto out;
147 	}
148 	map->m_la = chunknr << vi->chunkbits;
149 	map->m_plen = min_t(erofs_off_t, 1UL << vi->chunkbits,
150 			round_up(inode->i_size - map->m_la, sb->s_blocksize));
151 
152 	/* handle block map */
153 	if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
154 		__le32 *blkaddr = kaddr;
155 
156 		if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
157 			map->m_flags = 0;
158 		} else {
159 			map->m_pa = erofs_pos(sb, le32_to_cpu(*blkaddr));
160 			map->m_flags = EROFS_MAP_MAPPED;
161 		}
162 		goto out_unlock;
163 	}
164 	/* parse chunk indexes */
165 	idx = kaddr;
166 	switch (le32_to_cpu(idx->blkaddr)) {
167 	case EROFS_NULL_ADDR:
168 		map->m_flags = 0;
169 		break;
170 	default:
171 		map->m_deviceid = le16_to_cpu(idx->device_id) &
172 			EROFS_SB(sb)->device_id_mask;
173 		map->m_pa = erofs_pos(sb, le32_to_cpu(idx->blkaddr));
174 		map->m_flags = EROFS_MAP_MAPPED;
175 		break;
176 	}
177 out_unlock:
178 	erofs_put_metabuf(&buf);
179 out:
180 	if (!err)
181 		map->m_llen = map->m_plen;
182 	trace_erofs_map_blocks_exit(inode, map, 0, err);
183 	return err;
184 }
185 
186 static void erofs_fill_from_devinfo(struct erofs_map_dev *map,
187 				    struct erofs_device_info *dif)
188 {
189 	map->m_bdev = NULL;
190 	map->m_fp = NULL;
191 	if (dif->file) {
192 		if (S_ISBLK(file_inode(dif->file)->i_mode))
193 			map->m_bdev = file_bdev(dif->file);
194 		else
195 			map->m_fp = dif->file;
196 	}
197 	map->m_daxdev = dif->dax_dev;
198 	map->m_dax_part_off = dif->dax_part_off;
199 	map->m_fscache = dif->fscache;
200 }
201 
202 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
203 {
204 	struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
205 	struct erofs_device_info *dif;
206 	erofs_off_t startoff, length;
207 	int id;
208 
209 	map->m_bdev = sb->s_bdev;
210 	map->m_daxdev = EROFS_SB(sb)->dax_dev;
211 	map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
212 	map->m_fscache = EROFS_SB(sb)->s_fscache;
213 	map->m_fp = EROFS_SB(sb)->fdev;
214 
215 	if (map->m_deviceid) {
216 		down_read(&devs->rwsem);
217 		dif = idr_find(&devs->tree, map->m_deviceid - 1);
218 		if (!dif) {
219 			up_read(&devs->rwsem);
220 			return -ENODEV;
221 		}
222 		if (devs->flatdev) {
223 			map->m_pa += erofs_pos(sb, dif->mapped_blkaddr);
224 			up_read(&devs->rwsem);
225 			return 0;
226 		}
227 		erofs_fill_from_devinfo(map, dif);
228 		up_read(&devs->rwsem);
229 	} else if (devs->extra_devices && !devs->flatdev) {
230 		down_read(&devs->rwsem);
231 		idr_for_each_entry(&devs->tree, dif, id) {
232 			if (!dif->mapped_blkaddr)
233 				continue;
234 
235 			startoff = erofs_pos(sb, dif->mapped_blkaddr);
236 			length = erofs_pos(sb, dif->blocks);
237 			if (map->m_pa >= startoff &&
238 			    map->m_pa < startoff + length) {
239 				map->m_pa -= startoff;
240 				erofs_fill_from_devinfo(map, dif);
241 				break;
242 			}
243 		}
244 		up_read(&devs->rwsem);
245 	}
246 	return 0;
247 }
248 
249 /*
250  * bit 30: I/O error occurred on this folio
251  * bit 0 - 29: remaining parts to complete this folio
252  */
253 #define EROFS_ONLINEFOLIO_EIO			(1 << 30)
254 
255 void erofs_onlinefolio_init(struct folio *folio)
256 {
257 	union {
258 		atomic_t o;
259 		void *v;
260 	} u = { .o = ATOMIC_INIT(1) };
261 
262 	folio->private = u.v;	/* valid only if file-backed folio is locked */
263 }
264 
265 void erofs_onlinefolio_split(struct folio *folio)
266 {
267 	atomic_inc((atomic_t *)&folio->private);
268 }
269 
270 void erofs_onlinefolio_end(struct folio *folio, int err)
271 {
272 	int orig, v;
273 
274 	do {
275 		orig = atomic_read((atomic_t *)&folio->private);
276 		v = (orig - 1) | (err ? EROFS_ONLINEFOLIO_EIO : 0);
277 	} while (atomic_cmpxchg((atomic_t *)&folio->private, orig, v) != orig);
278 
279 	if (v & ~EROFS_ONLINEFOLIO_EIO)
280 		return;
281 	folio->private = 0;
282 	folio_end_read(folio, !(v & EROFS_ONLINEFOLIO_EIO));
283 }
284 
285 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
286 		unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
287 {
288 	int ret;
289 	struct super_block *sb = inode->i_sb;
290 	struct erofs_map_blocks map;
291 	struct erofs_map_dev mdev;
292 
293 	map.m_la = offset;
294 	map.m_llen = length;
295 
296 	ret = erofs_map_blocks(inode, &map);
297 	if (ret < 0)
298 		return ret;
299 
300 	mdev = (struct erofs_map_dev) {
301 		.m_deviceid = map.m_deviceid,
302 		.m_pa = map.m_pa,
303 	};
304 	ret = erofs_map_dev(sb, &mdev);
305 	if (ret)
306 		return ret;
307 
308 	iomap->offset = map.m_la;
309 	if (flags & IOMAP_DAX)
310 		iomap->dax_dev = mdev.m_daxdev;
311 	else
312 		iomap->bdev = mdev.m_bdev;
313 	iomap->length = map.m_llen;
314 	iomap->flags = 0;
315 	iomap->private = NULL;
316 
317 	if (!(map.m_flags & EROFS_MAP_MAPPED)) {
318 		iomap->type = IOMAP_HOLE;
319 		iomap->addr = IOMAP_NULL_ADDR;
320 		if (!iomap->length)
321 			iomap->length = length;
322 		return 0;
323 	}
324 
325 	if (map.m_flags & EROFS_MAP_META) {
326 		void *ptr;
327 		struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
328 
329 		iomap->type = IOMAP_INLINE;
330 		ptr = erofs_read_metabuf(&buf, sb, mdev.m_pa, EROFS_KMAP);
331 		if (IS_ERR(ptr))
332 			return PTR_ERR(ptr);
333 		iomap->inline_data = ptr;
334 		iomap->private = buf.base;
335 	} else {
336 		iomap->type = IOMAP_MAPPED;
337 		iomap->addr = mdev.m_pa;
338 		if (flags & IOMAP_DAX)
339 			iomap->addr += mdev.m_dax_part_off;
340 	}
341 	return 0;
342 }
343 
344 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
345 		ssize_t written, unsigned int flags, struct iomap *iomap)
346 {
347 	void *ptr = iomap->private;
348 
349 	if (ptr) {
350 		struct erofs_buf buf = {
351 			.page = kmap_to_page(ptr),
352 			.base = ptr,
353 			.kmap_type = EROFS_KMAP,
354 		};
355 
356 		DBG_BUGON(iomap->type != IOMAP_INLINE);
357 		erofs_put_metabuf(&buf);
358 	} else {
359 		DBG_BUGON(iomap->type == IOMAP_INLINE);
360 	}
361 	return written;
362 }
363 
364 static const struct iomap_ops erofs_iomap_ops = {
365 	.iomap_begin = erofs_iomap_begin,
366 	.iomap_end = erofs_iomap_end,
367 };
368 
369 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
370 		 u64 start, u64 len)
371 {
372 	if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
373 #ifdef CONFIG_EROFS_FS_ZIP
374 		return iomap_fiemap(inode, fieinfo, start, len,
375 				    &z_erofs_iomap_report_ops);
376 #else
377 		return -EOPNOTSUPP;
378 #endif
379 	}
380 	return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops);
381 }
382 
383 /*
384  * since we dont have write or truncate flows, so no inode
385  * locking needs to be held at the moment.
386  */
387 static int erofs_read_folio(struct file *file, struct folio *folio)
388 {
389 	return iomap_read_folio(folio, &erofs_iomap_ops);
390 }
391 
392 static void erofs_readahead(struct readahead_control *rac)
393 {
394 	return iomap_readahead(rac, &erofs_iomap_ops);
395 }
396 
397 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
398 {
399 	return iomap_bmap(mapping, block, &erofs_iomap_ops);
400 }
401 
402 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
403 {
404 	struct inode *inode = file_inode(iocb->ki_filp);
405 
406 	/* no need taking (shared) inode lock since it's a ro filesystem */
407 	if (!iov_iter_count(to))
408 		return 0;
409 
410 #ifdef CONFIG_FS_DAX
411 	if (IS_DAX(inode))
412 		return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
413 #endif
414 	if (iocb->ki_flags & IOCB_DIRECT) {
415 		struct block_device *bdev = inode->i_sb->s_bdev;
416 		unsigned int blksize_mask;
417 
418 		if (bdev)
419 			blksize_mask = bdev_logical_block_size(bdev) - 1;
420 		else
421 			blksize_mask = i_blocksize(inode) - 1;
422 
423 		if ((iocb->ki_pos | iov_iter_count(to) |
424 		     iov_iter_alignment(to)) & blksize_mask)
425 			return -EINVAL;
426 
427 		return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
428 				    NULL, 0, NULL, 0);
429 	}
430 	return filemap_read(iocb, to, 0);
431 }
432 
433 /* for uncompressed (aligned) files and raw access for other files */
434 const struct address_space_operations erofs_aops = {
435 	.read_folio = erofs_read_folio,
436 	.readahead = erofs_readahead,
437 	.bmap = erofs_bmap,
438 	.direct_IO = noop_direct_IO,
439 	.release_folio = iomap_release_folio,
440 	.invalidate_folio = iomap_invalidate_folio,
441 };
442 
443 #ifdef CONFIG_FS_DAX
444 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
445 		unsigned int order)
446 {
447 	return dax_iomap_fault(vmf, order, NULL, NULL, &erofs_iomap_ops);
448 }
449 
450 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
451 {
452 	return erofs_dax_huge_fault(vmf, 0);
453 }
454 
455 static const struct vm_operations_struct erofs_dax_vm_ops = {
456 	.fault		= erofs_dax_fault,
457 	.huge_fault	= erofs_dax_huge_fault,
458 };
459 
460 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
461 {
462 	if (!IS_DAX(file_inode(file)))
463 		return generic_file_readonly_mmap(file, vma);
464 
465 	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
466 		return -EINVAL;
467 
468 	vma->vm_ops = &erofs_dax_vm_ops;
469 	vm_flags_set(vma, VM_HUGEPAGE);
470 	return 0;
471 }
472 #else
473 #define erofs_file_mmap	generic_file_readonly_mmap
474 #endif
475 
476 const struct file_operations erofs_file_fops = {
477 	.llseek		= generic_file_llseek,
478 	.read_iter	= erofs_file_read_iter,
479 	.mmap		= erofs_file_mmap,
480 	.get_unmapped_area = thp_get_unmapped_area,
481 	.splice_read	= filemap_splice_read,
482 };
483