xref: /openbmc/linux/fs/erofs/data.c (revision e7f127b2)
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/prefetch.h>
9 #include <linux/dax.h>
10 #include <trace/events/erofs.h>
11 
12 void erofs_unmap_metabuf(struct erofs_buf *buf)
13 {
14 	if (buf->kmap_type == EROFS_KMAP)
15 		kunmap(buf->page);
16 	else if (buf->kmap_type == EROFS_KMAP_ATOMIC)
17 		kunmap_atomic(buf->base);
18 	buf->base = NULL;
19 	buf->kmap_type = EROFS_NO_KMAP;
20 }
21 
22 void erofs_put_metabuf(struct erofs_buf *buf)
23 {
24 	if (!buf->page)
25 		return;
26 	erofs_unmap_metabuf(buf);
27 	put_page(buf->page);
28 	buf->page = NULL;
29 }
30 
31 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
32 			erofs_blk_t blkaddr, enum erofs_kmap_type type)
33 {
34 	struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
35 	erofs_off_t offset = blknr_to_addr(blkaddr);
36 	pgoff_t index = offset >> PAGE_SHIFT;
37 	struct page *page = buf->page;
38 
39 	if (!page || page->index != index) {
40 		erofs_put_metabuf(buf);
41 		page = read_cache_page_gfp(mapping, index,
42 				mapping_gfp_constraint(mapping, ~__GFP_FS));
43 		if (IS_ERR(page))
44 			return page;
45 		/* should already be PageUptodate, no need to lock page */
46 		buf->page = page;
47 	}
48 	if (buf->kmap_type == EROFS_NO_KMAP) {
49 		if (type == EROFS_KMAP)
50 			buf->base = kmap(page);
51 		else if (type == EROFS_KMAP_ATOMIC)
52 			buf->base = kmap_atomic(page);
53 		buf->kmap_type = type;
54 	} else if (buf->kmap_type != type) {
55 		DBG_BUGON(1);
56 		return ERR_PTR(-EFAULT);
57 	}
58 	if (type == EROFS_NO_KMAP)
59 		return NULL;
60 	return buf->base + (offset & ~PAGE_MASK);
61 }
62 
63 static int erofs_map_blocks_flatmode(struct inode *inode,
64 				     struct erofs_map_blocks *map,
65 				     int flags)
66 {
67 	erofs_blk_t nblocks, lastblk;
68 	u64 offset = map->m_la;
69 	struct erofs_inode *vi = EROFS_I(inode);
70 	bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
71 
72 	nblocks = DIV_ROUND_UP(inode->i_size, EROFS_BLKSIZ);
73 	lastblk = nblocks - tailendpacking;
74 
75 	/* there is no hole in flatmode */
76 	map->m_flags = EROFS_MAP_MAPPED;
77 	if (offset < blknr_to_addr(lastblk)) {
78 		map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la;
79 		map->m_plen = blknr_to_addr(lastblk) - offset;
80 	} else if (tailendpacking) {
81 		/* 2 - inode inline B: inode, [xattrs], inline last blk... */
82 		struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
83 
84 		map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize +
85 			vi->xattr_isize + erofs_blkoff(map->m_la);
86 		map->m_plen = inode->i_size - offset;
87 
88 		/* inline data should be located in the same meta block */
89 		if (erofs_blkoff(map->m_pa) + map->m_plen > EROFS_BLKSIZ) {
90 			erofs_err(inode->i_sb,
91 				  "inline data cross block boundary @ nid %llu",
92 				  vi->nid);
93 			DBG_BUGON(1);
94 			return -EFSCORRUPTED;
95 		}
96 		map->m_flags |= EROFS_MAP_META;
97 	} else {
98 		erofs_err(inode->i_sb,
99 			  "internal error @ nid: %llu (size %llu), m_la 0x%llx",
100 			  vi->nid, inode->i_size, map->m_la);
101 		DBG_BUGON(1);
102 		return -EIO;
103 	}
104 	return 0;
105 }
106 
107 static int erofs_map_blocks(struct inode *inode,
108 			    struct erofs_map_blocks *map, int flags)
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, flags);
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 = 0;
126 		goto out;
127 	}
128 
129 	if (vi->datalayout != EROFS_INODE_CHUNK_BASED) {
130 		err = erofs_map_blocks_flatmode(inode, map, flags);
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(iloc(EROFS_SB(sb), vi->nid) + vi->inode_isize +
141 		    vi->xattr_isize, unit) + unit * chunknr;
142 
143 	kaddr = erofs_read_metabuf(&buf, sb, erofs_blknr(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 			    roundup(inode->i_size - map->m_la, EROFS_BLKSIZ));
151 
152 	/* handle block map */
153 	if (!(vi->chunkformat & EROFS_CHUNK_FORMAT_INDEXES)) {
154 		__le32 *blkaddr = kaddr + erofs_blkoff(pos);
155 
156 		if (le32_to_cpu(*blkaddr) == EROFS_NULL_ADDR) {
157 			map->m_flags = 0;
158 		} else {
159 			map->m_pa = blknr_to_addr(le32_to_cpu(*blkaddr));
160 			map->m_flags = EROFS_MAP_MAPPED;
161 		}
162 		goto out_unlock;
163 	}
164 	/* parse chunk indexes */
165 	idx = kaddr + erofs_blkoff(pos);
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 = blknr_to_addr(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, flags, 0);
183 	return err;
184 }
185 
186 int erofs_map_dev(struct super_block *sb, struct erofs_map_dev *map)
187 {
188 	struct erofs_dev_context *devs = EROFS_SB(sb)->devs;
189 	struct erofs_device_info *dif;
190 	int id;
191 
192 	/* primary device by default */
193 	map->m_bdev = sb->s_bdev;
194 	map->m_daxdev = EROFS_SB(sb)->dax_dev;
195 	map->m_dax_part_off = EROFS_SB(sb)->dax_part_off;
196 
197 	if (map->m_deviceid) {
198 		down_read(&devs->rwsem);
199 		dif = idr_find(&devs->tree, map->m_deviceid - 1);
200 		if (!dif) {
201 			up_read(&devs->rwsem);
202 			return -ENODEV;
203 		}
204 		map->m_bdev = dif->bdev;
205 		map->m_daxdev = dif->dax_dev;
206 		map->m_dax_part_off = dif->dax_part_off;
207 		up_read(&devs->rwsem);
208 	} else if (devs->extra_devices) {
209 		down_read(&devs->rwsem);
210 		idr_for_each_entry(&devs->tree, dif, id) {
211 			erofs_off_t startoff, length;
212 
213 			if (!dif->mapped_blkaddr)
214 				continue;
215 			startoff = blknr_to_addr(dif->mapped_blkaddr);
216 			length = blknr_to_addr(dif->blocks);
217 
218 			if (map->m_pa >= startoff &&
219 			    map->m_pa < startoff + length) {
220 				map->m_pa -= startoff;
221 				map->m_bdev = dif->bdev;
222 				map->m_daxdev = dif->dax_dev;
223 				map->m_dax_part_off = dif->dax_part_off;
224 				break;
225 			}
226 		}
227 		up_read(&devs->rwsem);
228 	}
229 	return 0;
230 }
231 
232 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
233 		unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
234 {
235 	int ret;
236 	struct erofs_map_blocks map;
237 	struct erofs_map_dev mdev;
238 
239 	map.m_la = offset;
240 	map.m_llen = length;
241 
242 	ret = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
243 	if (ret < 0)
244 		return ret;
245 
246 	mdev = (struct erofs_map_dev) {
247 		.m_deviceid = map.m_deviceid,
248 		.m_pa = map.m_pa,
249 	};
250 	ret = erofs_map_dev(inode->i_sb, &mdev);
251 	if (ret)
252 		return ret;
253 
254 	iomap->offset = map.m_la;
255 	if (flags & IOMAP_DAX) {
256 		iomap->dax_dev = mdev.m_daxdev;
257 		iomap->offset += mdev.m_dax_part_off;
258 	} else {
259 		iomap->bdev = mdev.m_bdev;
260 	}
261 	iomap->length = map.m_llen;
262 	iomap->flags = 0;
263 	iomap->private = NULL;
264 
265 	if (!(map.m_flags & EROFS_MAP_MAPPED)) {
266 		iomap->type = IOMAP_HOLE;
267 		iomap->addr = IOMAP_NULL_ADDR;
268 		if (!iomap->length)
269 			iomap->length = length;
270 		return 0;
271 	}
272 
273 	if (map.m_flags & EROFS_MAP_META) {
274 		void *ptr;
275 		struct erofs_buf buf = __EROFS_BUF_INITIALIZER;
276 
277 		iomap->type = IOMAP_INLINE;
278 		ptr = erofs_read_metabuf(&buf, inode->i_sb,
279 					 erofs_blknr(mdev.m_pa), EROFS_KMAP);
280 		if (IS_ERR(ptr))
281 			return PTR_ERR(ptr);
282 		iomap->inline_data = ptr + erofs_blkoff(mdev.m_pa);
283 		iomap->private = buf.base;
284 	} else {
285 		iomap->type = IOMAP_MAPPED;
286 		iomap->addr = mdev.m_pa;
287 	}
288 	return 0;
289 }
290 
291 static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
292 		ssize_t written, unsigned int flags, struct iomap *iomap)
293 {
294 	void *ptr = iomap->private;
295 
296 	if (ptr) {
297 		struct erofs_buf buf = {
298 			.page = kmap_to_page(ptr),
299 			.base = ptr,
300 			.kmap_type = EROFS_KMAP,
301 		};
302 
303 		DBG_BUGON(iomap->type != IOMAP_INLINE);
304 		erofs_put_metabuf(&buf);
305 	} else {
306 		DBG_BUGON(iomap->type == IOMAP_INLINE);
307 	}
308 	return written;
309 }
310 
311 static const struct iomap_ops erofs_iomap_ops = {
312 	.iomap_begin = erofs_iomap_begin,
313 	.iomap_end = erofs_iomap_end,
314 };
315 
316 int erofs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
317 		 u64 start, u64 len)
318 {
319 	if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
320 #ifdef CONFIG_EROFS_FS_ZIP
321 		return iomap_fiemap(inode, fieinfo, start, len,
322 				    &z_erofs_iomap_report_ops);
323 #else
324 		return -EOPNOTSUPP;
325 #endif
326 	}
327 	return iomap_fiemap(inode, fieinfo, start, len, &erofs_iomap_ops);
328 }
329 
330 /*
331  * since we dont have write or truncate flows, so no inode
332  * locking needs to be held at the moment.
333  */
334 static int erofs_readpage(struct file *file, struct page *page)
335 {
336 	return iomap_readpage(page, &erofs_iomap_ops);
337 }
338 
339 static void erofs_readahead(struct readahead_control *rac)
340 {
341 	return iomap_readahead(rac, &erofs_iomap_ops);
342 }
343 
344 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
345 {
346 	return iomap_bmap(mapping, block, &erofs_iomap_ops);
347 }
348 
349 static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to)
350 {
351 	struct inode *inode = file_inode(iocb->ki_filp);
352 	loff_t align = iocb->ki_pos | iov_iter_count(to) |
353 		iov_iter_alignment(to);
354 	struct block_device *bdev = inode->i_sb->s_bdev;
355 	unsigned int blksize_mask;
356 
357 	if (bdev)
358 		blksize_mask = (1 << ilog2(bdev_logical_block_size(bdev))) - 1;
359 	else
360 		blksize_mask = (1 << inode->i_blkbits) - 1;
361 
362 	if (align & blksize_mask)
363 		return -EINVAL;
364 	return 0;
365 }
366 
367 static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
368 {
369 	/* no need taking (shared) inode lock since it's a ro filesystem */
370 	if (!iov_iter_count(to))
371 		return 0;
372 
373 #ifdef CONFIG_FS_DAX
374 	if (IS_DAX(iocb->ki_filp->f_mapping->host))
375 		return dax_iomap_rw(iocb, to, &erofs_iomap_ops);
376 #endif
377 	if (iocb->ki_flags & IOCB_DIRECT) {
378 		int err = erofs_prepare_dio(iocb, to);
379 
380 		if (!err)
381 			return iomap_dio_rw(iocb, to, &erofs_iomap_ops,
382 					    NULL, 0, 0);
383 		if (err < 0)
384 			return err;
385 	}
386 	return filemap_read(iocb, to, 0);
387 }
388 
389 /* for uncompressed (aligned) files and raw access for other files */
390 const struct address_space_operations erofs_raw_access_aops = {
391 	.readpage = erofs_readpage,
392 	.readahead = erofs_readahead,
393 	.bmap = erofs_bmap,
394 	.direct_IO = noop_direct_IO,
395 };
396 
397 #ifdef CONFIG_FS_DAX
398 static vm_fault_t erofs_dax_huge_fault(struct vm_fault *vmf,
399 		enum page_entry_size pe_size)
400 {
401 	return dax_iomap_fault(vmf, pe_size, NULL, NULL, &erofs_iomap_ops);
402 }
403 
404 static vm_fault_t erofs_dax_fault(struct vm_fault *vmf)
405 {
406 	return erofs_dax_huge_fault(vmf, PE_SIZE_PTE);
407 }
408 
409 static const struct vm_operations_struct erofs_dax_vm_ops = {
410 	.fault		= erofs_dax_fault,
411 	.huge_fault	= erofs_dax_huge_fault,
412 };
413 
414 static int erofs_file_mmap(struct file *file, struct vm_area_struct *vma)
415 {
416 	if (!IS_DAX(file_inode(file)))
417 		return generic_file_readonly_mmap(file, vma);
418 
419 	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
420 		return -EINVAL;
421 
422 	vma->vm_ops = &erofs_dax_vm_ops;
423 	vma->vm_flags |= VM_HUGEPAGE;
424 	return 0;
425 }
426 #else
427 #define erofs_file_mmap	generic_file_readonly_mmap
428 #endif
429 
430 const struct file_operations erofs_file_fops = {
431 	.llseek		= generic_file_llseek,
432 	.read_iter	= erofs_file_read_iter,
433 	.mmap		= erofs_file_mmap,
434 	.splice_read	= generic_file_splice_read,
435 };
436