1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/drivers/staging/erofs/data.c
4  *
5  * Copyright (C) 2017-2018 HUAWEI, Inc.
6  *             http://www.huawei.com/
7  * Created by Gao Xiang <gaoxiang25@huawei.com>
8  *
9  * This file is subject to the terms and conditions of the GNU General Public
10  * License.  See the file COPYING in the main directory of the Linux
11  * distribution for more details.
12  */
13 #include "internal.h"
14 #include <linux/prefetch.h>
15 
16 #include <trace/events/erofs.h>
17 
read_endio(struct bio * bio)18 static inline void read_endio(struct bio *bio)
19 {
20 	int i;
21 	struct bio_vec *bvec;
22 	const blk_status_t err = bio->bi_status;
23 
24 	bio_for_each_segment_all(bvec, bio, i) {
25 		struct page *page = bvec->bv_page;
26 
27 		/* page is already locked */
28 		BUG_ON(PageUptodate(page));
29 
30 		if (unlikely(err))
31 			SetPageError(page);
32 		else
33 			SetPageUptodate(page);
34 
35 		unlock_page(page);
36 		/* page could be reclaimed now */
37 	}
38 	bio_put(bio);
39 }
40 
41 /* prio -- true is used for dir */
erofs_get_meta_page(struct super_block * sb,erofs_blk_t blkaddr,bool prio)42 struct page *erofs_get_meta_page(struct super_block *sb,
43 	erofs_blk_t blkaddr, bool prio)
44 {
45 	struct inode *bd_inode = sb->s_bdev->bd_inode;
46 	struct address_space *mapping = bd_inode->i_mapping;
47 	struct page *page;
48 
49 repeat:
50 	page = find_or_create_page(mapping, blkaddr,
51 	/*
52 	 * Prefer looping in the allocator rather than here,
53 	 * at least that code knows what it's doing.
54 	 */
55 		mapping_gfp_constraint(mapping, ~__GFP_FS) | __GFP_NOFAIL);
56 
57 	BUG_ON(!page || !PageLocked(page));
58 
59 	if (!PageUptodate(page)) {
60 		struct bio *bio;
61 		int err;
62 
63 		bio = prepare_bio(sb, blkaddr, 1, read_endio);
64 		err = bio_add_page(bio, page, PAGE_SIZE, 0);
65 		BUG_ON(err != PAGE_SIZE);
66 
67 		__submit_bio(bio, REQ_OP_READ,
68 			REQ_META | (prio ? REQ_PRIO : 0));
69 
70 		lock_page(page);
71 
72 		/* the page has been truncated by others? */
73 		if (unlikely(page->mapping != mapping)) {
74 			unlock_page(page);
75 			put_page(page);
76 			goto repeat;
77 		}
78 
79 		/* more likely a read error */
80 		if (unlikely(!PageUptodate(page))) {
81 			unlock_page(page);
82 			put_page(page);
83 
84 			page = ERR_PTR(-EIO);
85 		}
86 	}
87 	return page;
88 }
89 
erofs_map_blocks_flatmode(struct inode * inode,struct erofs_map_blocks * map,int flags)90 static int erofs_map_blocks_flatmode(struct inode *inode,
91 	struct erofs_map_blocks *map,
92 	int flags)
93 {
94 	erofs_blk_t nblocks, lastblk;
95 	u64 offset = map->m_la;
96 	struct erofs_vnode *vi = EROFS_V(inode);
97 
98 	trace_erofs_map_blocks_flatmode_enter(inode, map, flags);
99 	BUG_ON(is_inode_layout_compression(inode));
100 
101 	nblocks = DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
102 	lastblk = nblocks - is_inode_layout_inline(inode);
103 
104 	if (unlikely(offset >= inode->i_size)) {
105 		/* leave out-of-bound access unmapped */
106 		map->m_flags = 0;
107 		map->m_plen = 0;
108 		goto out;
109 	}
110 
111 	/* there is no hole in flatmode */
112 	map->m_flags = EROFS_MAP_MAPPED;
113 
114 	if (offset < blknr_to_addr(lastblk)) {
115 		map->m_pa = blknr_to_addr(vi->raw_blkaddr) + map->m_la;
116 		map->m_plen = blknr_to_addr(lastblk) - offset;
117 	} else if (is_inode_layout_inline(inode)) {
118 		/* 2 - inode inline B: inode, [xattrs], inline last blk... */
119 		struct erofs_sb_info *sbi = EROFS_SB(inode->i_sb);
120 
121 		map->m_pa = iloc(sbi, vi->nid) + vi->inode_isize +
122 			vi->xattr_isize + erofs_blkoff(map->m_la);
123 		map->m_plen = inode->i_size - offset;
124 
125 		/* inline data should locate in one meta block */
126 		BUG_ON(erofs_blkoff(map->m_pa) + map->m_plen > PAGE_SIZE);
127 		map->m_flags |= EROFS_MAP_META;
128 	} else {
129 		errln("internal error @ nid: %llu (size %llu), m_la 0x%llx",
130 			vi->nid, inode->i_size, map->m_la);
131 		BUG();
132 	}
133 
134 out:
135 	map->m_llen = map->m_plen;
136 	trace_erofs_map_blocks_flatmode_exit(inode, map, flags, 0);
137 	return 0;
138 }
139 
140 #ifdef CONFIG_EROFS_FS_ZIP
141 extern int z_erofs_map_blocks_iter(struct inode *,
142 	struct erofs_map_blocks *, struct page **, int);
143 #endif
144 
erofs_map_blocks_iter(struct inode * inode,struct erofs_map_blocks * map,struct page ** mpage_ret,int flags)145 int erofs_map_blocks_iter(struct inode *inode,
146 	struct erofs_map_blocks *map,
147 	struct page **mpage_ret, int flags)
148 {
149 	/* by default, reading raw data never use erofs_map_blocks_iter */
150 	if (unlikely(!is_inode_layout_compression(inode))) {
151 		if (*mpage_ret != NULL)
152 			put_page(*mpage_ret);
153 		*mpage_ret = NULL;
154 
155 		return erofs_map_blocks(inode, map, flags);
156 	}
157 
158 #ifdef CONFIG_EROFS_FS_ZIP
159 	return z_erofs_map_blocks_iter(inode, map, mpage_ret, flags);
160 #else
161 	/* data compression is not available */
162 	return -ENOTSUPP;
163 #endif
164 }
165 
erofs_map_blocks(struct inode * inode,struct erofs_map_blocks * map,int flags)166 int erofs_map_blocks(struct inode *inode,
167 	struct erofs_map_blocks *map, int flags)
168 {
169 	if (unlikely(is_inode_layout_compression(inode))) {
170 		struct page *mpage = NULL;
171 		int err;
172 
173 		err = erofs_map_blocks_iter(inode, map, &mpage, flags);
174 		if (mpage != NULL)
175 			put_page(mpage);
176 		return err;
177 	}
178 	return erofs_map_blocks_flatmode(inode, map, flags);
179 }
180 
erofs_read_raw_page(struct bio * bio,struct address_space * mapping,struct page * page,erofs_off_t * last_block,unsigned nblocks,bool ra)181 static inline struct bio *erofs_read_raw_page(
182 	struct bio *bio,
183 	struct address_space *mapping,
184 	struct page *page,
185 	erofs_off_t *last_block,
186 	unsigned nblocks,
187 	bool ra)
188 {
189 	struct inode *inode = mapping->host;
190 	erofs_off_t current_block = (erofs_off_t)page->index;
191 	int err;
192 
193 	BUG_ON(!nblocks);
194 
195 	if (PageUptodate(page)) {
196 		err = 0;
197 		goto has_updated;
198 	}
199 
200 	if (cleancache_get_page(page) == 0) {
201 		err = 0;
202 		SetPageUptodate(page);
203 		goto has_updated;
204 	}
205 
206 	/* note that for readpage case, bio also equals to NULL */
207 	if (bio != NULL &&
208 		/* not continuous */
209 		*last_block + 1 != current_block) {
210 submit_bio_retry:
211 		__submit_bio(bio, REQ_OP_READ, 0);
212 		bio = NULL;
213 	}
214 
215 	if (bio == NULL) {
216 		struct erofs_map_blocks map = {
217 			.m_la = blknr_to_addr(current_block),
218 		};
219 		erofs_blk_t blknr;
220 		unsigned blkoff;
221 
222 		err = erofs_map_blocks(inode, &map, EROFS_GET_BLOCKS_RAW);
223 		if (unlikely(err))
224 			goto err_out;
225 
226 		/* zero out the holed page */
227 		if (unlikely(!(map.m_flags & EROFS_MAP_MAPPED))) {
228 			zero_user_segment(page, 0, PAGE_SIZE);
229 			SetPageUptodate(page);
230 
231 			/* imply err = 0, see erofs_map_blocks */
232 			goto has_updated;
233 		}
234 
235 		/* for RAW access mode, m_plen must be equal to m_llen */
236 		BUG_ON(map.m_plen != map.m_llen);
237 
238 		blknr = erofs_blknr(map.m_pa);
239 		blkoff = erofs_blkoff(map.m_pa);
240 
241 		/* deal with inline page */
242 		if (map.m_flags & EROFS_MAP_META) {
243 			void *vsrc, *vto;
244 			struct page *ipage;
245 
246 			BUG_ON(map.m_plen > PAGE_SIZE);
247 
248 			ipage = erofs_get_meta_page(inode->i_sb, blknr, 0);
249 
250 			if (IS_ERR(ipage)) {
251 				err = PTR_ERR(ipage);
252 				goto err_out;
253 			}
254 
255 			vsrc = kmap_atomic(ipage);
256 			vto = kmap_atomic(page);
257 			memcpy(vto, vsrc + blkoff, map.m_plen);
258 			memset(vto + map.m_plen, 0, PAGE_SIZE - map.m_plen);
259 			kunmap_atomic(vto);
260 			kunmap_atomic(vsrc);
261 			flush_dcache_page(page);
262 
263 			SetPageUptodate(page);
264 			/* TODO: could we unlock the page earlier? */
265 			unlock_page(ipage);
266 			put_page(ipage);
267 
268 			/* imply err = 0, see erofs_map_blocks */
269 			goto has_updated;
270 		}
271 
272 		/* pa must be block-aligned for raw reading */
273 		BUG_ON(erofs_blkoff(map.m_pa) != 0);
274 
275 		/* max # of continuous pages */
276 		if (nblocks > DIV_ROUND_UP(map.m_plen, PAGE_SIZE))
277 			nblocks = DIV_ROUND_UP(map.m_plen, PAGE_SIZE);
278 		if (nblocks > BIO_MAX_PAGES)
279 			nblocks = BIO_MAX_PAGES;
280 
281 		bio = prepare_bio(inode->i_sb, blknr, nblocks, read_endio);
282 	}
283 
284 	err = bio_add_page(bio, page, PAGE_SIZE, 0);
285 	/* out of the extent or bio is full */
286 	if (err < PAGE_SIZE)
287 		goto submit_bio_retry;
288 
289 	*last_block = current_block;
290 
291 	/* shift in advance in case of it followed by too many gaps */
292 	if (unlikely(bio->bi_vcnt >= bio->bi_max_vecs)) {
293 		/* err should reassign to 0 after submitting */
294 		err = 0;
295 		goto submit_bio_out;
296 	}
297 
298 	return bio;
299 
300 err_out:
301 	/* for sync reading, set page error immediately */
302 	if (!ra) {
303 		SetPageError(page);
304 		ClearPageUptodate(page);
305 	}
306 has_updated:
307 	unlock_page(page);
308 
309 	/* if updated manually, continuous pages has a gap */
310 	if (bio != NULL)
311 submit_bio_out:
312 		__submit_bio(bio, REQ_OP_READ, 0);
313 
314 	return unlikely(err) ? ERR_PTR(err) : NULL;
315 }
316 
317 /*
318  * since we dont have write or truncate flows, so no inode
319  * locking needs to be held at the moment.
320  */
erofs_raw_access_readpage(struct file * file,struct page * page)321 static int erofs_raw_access_readpage(struct file *file, struct page *page)
322 {
323 	erofs_off_t last_block;
324 	struct bio *bio;
325 
326 	trace_erofs_readpage(page, true);
327 
328 	bio = erofs_read_raw_page(NULL, page->mapping,
329 		page, &last_block, 1, false);
330 
331 	if (IS_ERR(bio))
332 		return PTR_ERR(bio);
333 
334 	BUG_ON(bio != NULL);	/* since we have only one bio -- must be NULL */
335 	return 0;
336 }
337 
erofs_raw_access_readpages(struct file * filp,struct address_space * mapping,struct list_head * pages,unsigned nr_pages)338 static int erofs_raw_access_readpages(struct file *filp,
339 	struct address_space *mapping,
340 	struct list_head *pages, unsigned nr_pages)
341 {
342 	erofs_off_t last_block;
343 	struct bio *bio = NULL;
344 	gfp_t gfp = readahead_gfp_mask(mapping);
345 	struct page *page = list_last_entry(pages, struct page, lru);
346 
347 	trace_erofs_readpages(mapping->host, page, nr_pages, true);
348 
349 	for (; nr_pages; --nr_pages) {
350 		page = list_entry(pages->prev, struct page, lru);
351 
352 		prefetchw(&page->flags);
353 		list_del(&page->lru);
354 
355 		if (!add_to_page_cache_lru(page, mapping, page->index, gfp)) {
356 			bio = erofs_read_raw_page(bio, mapping, page,
357 				&last_block, nr_pages, true);
358 
359 			/* all the page errors are ignored when readahead */
360 			if (IS_ERR(bio)) {
361 				pr_err("%s, readahead error at page %lu of nid %llu\n",
362 					__func__, page->index,
363 					EROFS_V(mapping->host)->nid);
364 
365 				bio = NULL;
366 			}
367 		}
368 
369 		/* pages could still be locked */
370 		put_page(page);
371 	}
372 	BUG_ON(!list_empty(pages));
373 
374 	/* the rare case (end in gaps) */
375 	if (unlikely(bio != NULL))
376 		__submit_bio(bio, REQ_OP_READ, 0);
377 	return 0;
378 }
379 
380 /* for uncompressed (aligned) files and raw access for other files */
381 const struct address_space_operations erofs_raw_access_aops = {
382 	.readpage = erofs_raw_access_readpage,
383 	.readpages = erofs_raw_access_readpages,
384 };
385 
386