xref: /openbmc/linux/fs/erofs/data.c (revision 5f7136db) 
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017-2018 HUAWEI, Inc.
4  *             https://www.huawei.com/
5  * Created by Gao Xiang <gaoxiang25@huawei.com>
6  */
7 #include "internal.h"
8 #include <linux/prefetch.h>
9 
10 #include <trace/events/erofs.h>
11 
12 static void erofs_readendio(struct bio *bio)
13 {
14 	struct bio_vec *bvec;
15 	blk_status_t err = bio->bi_status;
16 	struct bvec_iter_all iter_all;
17 
18 	bio_for_each_segment_all(bvec, bio, iter_all) {
19 		struct page *page = bvec->bv_page;
20 
21 		/* page is already locked */
22 		DBG_BUGON(PageUptodate(page));
23 
24 		if (err)
25 			SetPageError(page);
26 		else
27 			SetPageUptodate(page);
28 
29 		unlock_page(page);
30 		/* page could be reclaimed now */
31 	}
32 	bio_put(bio);
33 }
34 
35 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
36 {
37 	struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
38 	struct page *page;
39 
40 	page = read_cache_page_gfp(mapping, blkaddr,
41 				   mapping_gfp_constraint(mapping, ~__GFP_FS));
42 	/* should already be PageUptodate */
43 	if (!IS_ERR(page))
44 		lock_page(page);
45 	return page;
46 }
47 
48 static int erofs_map_blocks_flatmode(struct inode *inode,
49 				     struct erofs_map_blocks *map,
50 				     int flags)
51 {
52 	int err = 0;
53 	erofs_blk_t nblocks, lastblk;
54 	u64 offset = map->m_la;
55 	struct erofs_inode *vi = EROFS_I(inode);
56 	bool tailendpacking = (vi->datalayout == EROFS_INODE_FLAT_INLINE);
57 
58 	trace_erofs_map_blocks_flatmode_enter(inode, map, flags);
59 
60 	nblocks = DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
61 	lastblk = nblocks - tailendpacking;
62 
63 	if (offset >= inode->i_size) {
64 		/* leave out-of-bound access unmapped */
65 		map->m_flags = 0;
66 		map->m_plen = 0;
67 		goto out;
68 	}
69 
70 	/* there is no hole in flatmode */
71 	map->m_flags = EROFS_MAP_MAPPED;
72 
73 	if (offset < blknr_to_addr(lastblk)) {
74 		map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la;
75 		map->m_plen = blknr_to_addr(lastblk) - offset;
76 	} else if (tailendpacking) {
77 		/* 2 - inode inline B: inode, [xattrs], inline last blk... */
78 		struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
79 
80 		map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize +
81 			vi->xattr_isize + erofs_blkoff(map->m_la);
82 		map->m_plen = inode->i_size - offset;
83 
84 		/* inline data should be located in one meta block */
85 		if (erofs_blkoff(map->m_pa) + map->m_plen > PAGE_SIZE) {
86 			erofs_err(inode->i_sb,
87 				  "inline data cross block boundary @ nid %llu",
88 				  vi->nid);
89 			DBG_BUGON(1);
90 			err = -EFSCORRUPTED;
91 			goto err_out;
92 		}
93 
94 		map->m_flags |= EROFS_MAP_META;
95 	} else {
96 		erofs_err(inode->i_sb,
97 			  "internal error @ nid: %llu (size %llu), m_la 0x%llx",
98 			  vi->nid, inode->i_size, map->m_la);
99 		DBG_BUGON(1);
100 		err = -EIO;
101 		goto err_out;
102 	}
103 
104 out:
105 	map->m_llen = map->m_plen;
106 
107 err_out:
108 	trace_erofs_map_blocks_flatmode_exit(inode, map, flags, 0);
109 	return err;
110 }
111 
112 int erofs_map_blocks(struct inode *inode,
113 		     struct erofs_map_blocks *map, int flags)
114 {
115 	if (erofs_inode_is_data_compressed(EROFS_I(inode)->datalayout)) {
116 		int err = z_erofs_map_blocks_iter(inode, map, flags);
117 
118 		if (map->mpage) {
119 			put_page(map->mpage);
120 			map->mpage = NULL;
121 		}
122 		return err;
123 	}
124 	return erofs_map_blocks_flatmode(inode, map, flags);
125 }
126 
127 static inline struct bio *erofs_read_raw_page(struct bio *bio,
128 					      struct address_space *mapping,
129 					      struct page *page,
130 					      erofs_off_t *last_block,
131 					      unsigned int nblocks,
132 					      bool ra)
133 {
134 	struct inode *const inode = mapping->host;
135 	struct super_block *const sb = inode->i_sb;
136 	erofs_off_t current_block = (erofs_off_t)page->index;
137 	int err;
138 
139 	DBG_BUGON(!nblocks);
140 
141 	if (PageUptodate(page)) {
142 		err = 0;
143 		goto has_updated;
144 	}
145 
146 	/* note that for readpage case, bio also equals to NULL */
147 	if (bio &&
148 	    /* not continuous */
149 	    *last_block + 1 != current_block) {
150 submit_bio_retry:
151 		submit_bio(bio);
152 		bio = NULL;
153 	}
154 
155 	if (!bio) {
156 		struct erofs_map_blocks map = {
157 			.m_la = blknr_to_addr(current_block),
158 		};
159 		erofs_blk_t blknr;
160 		unsigned int blkoff;
161 
162 		err = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
163 		if (err)
164 			goto err_out;
165 
166 		/* zero out the holed page */
167 		if (!(map.m_flags & EROFS_MAP_MAPPED)) {
168 			zero_user_segment(page, 0, PAGE_SIZE);
169 			SetPageUptodate(page);
170 
171 			/* imply err = 0, see erofs_map_blocks */
172 			goto has_updated;
173 		}
174 
175 		/* for RAW access mode, m_plen must be equal to m_llen */
176 		DBG_BUGON(map.m_plen != map.m_llen);
177 
178 		blknr = erofs_blknr(map.m_pa);
179 		blkoff = erofs_blkoff(map.m_pa);
180 
181 		/* deal with inline page */
182 		if (map.m_flags & EROFS_MAP_META) {
183 			void *vsrc, *vto;
184 			struct page *ipage;
185 
186 			DBG_BUGON(map.m_plen > PAGE_SIZE);
187 
188 			ipage = erofs_get_meta_page(inode->i_sb, blknr);
189 
190 			if (IS_ERR(ipage)) {
191 				err = PTR_ERR(ipage);
192 				goto err_out;
193 			}
194 
195 			vsrc = kmap_atomic(ipage);
196 			vto = kmap_atomic(page);
197 			memcpy(vto, vsrc + blkoff, map.m_plen);
198 			memset(vto + map.m_plen, 0, PAGE_SIZE - map.m_plen);
199 			kunmap_atomic(vto);
200 			kunmap_atomic(vsrc);
201 			flush_dcache_page(page);
202 
203 			SetPageUptodate(page);
204 			/* TODO: could we unlock the page earlier? */
205 			unlock_page(ipage);
206 			put_page(ipage);
207 
208 			/* imply err = 0, see erofs_map_blocks */
209 			goto has_updated;
210 		}
211 
212 		/* pa must be block-aligned for raw reading */
213 		DBG_BUGON(erofs_blkoff(map.m_pa));
214 
215 		/* max # of continuous pages */
216 		if (nblocks > DIV_ROUND_UP(map.m_plen, PAGE_SIZE))
217 			nblocks = DIV_ROUND_UP(map.m_plen, PAGE_SIZE);
218 
219 		bio = bio_alloc(GFP_NOIO, bio_max_segs(nblocks));
220 
221 		bio->bi_end_io = erofs_readendio;
222 		bio_set_dev(bio, sb->s_bdev);
223 		bio->bi_iter.bi_sector = (sector_t)blknr <<
224 			LOG_SECTORS_PER_BLOCK;
225 		bio->bi_opf = REQ_OP_READ | (ra ? REQ_RAHEAD : 0);
226 	}
227 
228 	err = bio_add_page(bio, page, PAGE_SIZE, 0);
229 	/* out of the extent or bio is full */
230 	if (err < PAGE_SIZE)
231 		goto submit_bio_retry;
232 
233 	*last_block = current_block;
234 
235 	/* shift in advance in case of it followed by too many gaps */
236 	if (bio->bi_iter.bi_size >= bio->bi_max_vecs * PAGE_SIZE) {
237 		/* err should reassign to 0 after submitting */
238 		err = 0;
239 		goto submit_bio_out;
240 	}
241 
242 	return bio;
243 
244 err_out:
245 	/* for sync reading, set page error immediately */
246 	if (!ra) {
247 		SetPageError(page);
248 		ClearPageUptodate(page);
249 	}
250 has_updated:
251 	unlock_page(page);
252 
253 	/* if updated manually, continuous pages has a gap */
254 	if (bio)
255 submit_bio_out:
256 		submit_bio(bio);
257 	return err ? ERR_PTR(err) : NULL;
258 }
259 
260 /*
261  * since we dont have write or truncate flows, so no inode
262  * locking needs to be held at the moment.
263  */
264 static int erofs_raw_access_readpage(struct file *file, struct page *page)
265 {
266 	erofs_off_t last_block;
267 	struct bio *bio;
268 
269 	trace_erofs_readpage(page, true);
270 
271 	bio = erofs_read_raw_page(NULL, page->mapping,
272 				  page, &last_block, 1, false);
273 
274 	if (IS_ERR(bio))
275 		return PTR_ERR(bio);
276 
277 	DBG_BUGON(bio);	/* since we have only one bio -- must be NULL */
278 	return 0;
279 }
280 
281 static void erofs_raw_access_readahead(struct readahead_control *rac)
282 {
283 	erofs_off_t last_block;
284 	struct bio *bio = NULL;
285 	struct page *page;
286 
287 	trace_erofs_readpages(rac->mapping->host, readahead_index(rac),
288 			readahead_count(rac), true);
289 
290 	while ((page = readahead_page(rac))) {
291 		prefetchw(&page->flags);
292 
293 		bio = erofs_read_raw_page(bio, rac->mapping, page, &last_block,
294 				readahead_count(rac), true);
295 
296 		/* all the page errors are ignored when readahead */
297 		if (IS_ERR(bio)) {
298 			pr_err("%s, readahead error at page %lu of nid %llu\n",
299 			       __func__, page->index,
300 			       EROFS_I(rac->mapping->host)->nid);
301 
302 			bio = NULL;
303 		}
304 
305 		put_page(page);
306 	}
307 
308 	/* the rare case (end in gaps) */
309 	if (bio)
310 		submit_bio(bio);
311 }
312 
313 static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
314 {
315 	struct inode *inode = mapping->host;
316 	struct erofs_map_blocks map = {
317 		.m_la = blknr_to_addr(block),
318 	};
319 
320 	if (EROFS_I(inode)->datalayout == EROFS_INODE_FLAT_INLINE) {
321 		erofs_blk_t blks = i_size_read(inode) >> LOG_BLOCK_SIZE;
322 
323 		if (block >> LOG_SECTORS_PER_BLOCK >= blks)
324 			return 0;
325 	}
326 
327 	if (!erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW))
328 		return erofs_blknr(map.m_pa);
329 
330 	return 0;
331 }
332 
333 /* for uncompressed (aligned) files and raw access for other files */
334 const struct address_space_operations erofs_raw_access_aops = {
335 	.readpage = erofs_raw_access_readpage,
336 	.readahead = erofs_raw_access_readahead,
337 	.bmap = erofs_bmap,
338 };
339 
340