xref: /openbmc/linux/fs/ext4/readpage.c (revision 4f3db074)
1 /*
2  * linux/fs/ext4/readpage.c
3  *
4  * Copyright (C) 2002, Linus Torvalds.
5  * Copyright (C) 2015, Google, Inc.
6  *
7  * This was originally taken from fs/mpage.c
8  *
9  * The intent is the ext4_mpage_readpages() function here is intended
10  * to replace mpage_readpages() in the general case, not just for
11  * encrypted files.  It has some limitations (see below), where it
12  * will fall back to read_block_full_page(), but these limitations
13  * should only be hit when page_size != block_size.
14  *
15  * This will allow us to attach a callback function to support ext4
16  * encryption.
17  *
18  * If anything unusual happens, such as:
19  *
20  * - encountering a page which has buffers
21  * - encountering a page which has a non-hole after a hole
22  * - encountering a page with non-contiguous blocks
23  *
24  * then this code just gives up and calls the buffer_head-based read function.
25  * It does handle a page which has holes at the end - that is a common case:
26  * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
27  *
28  */
29 
30 #include <linux/kernel.h>
31 #include <linux/export.h>
32 #include <linux/mm.h>
33 #include <linux/kdev_t.h>
34 #include <linux/gfp.h>
35 #include <linux/bio.h>
36 #include <linux/fs.h>
37 #include <linux/buffer_head.h>
38 #include <linux/blkdev.h>
39 #include <linux/highmem.h>
40 #include <linux/prefetch.h>
41 #include <linux/mpage.h>
42 #include <linux/writeback.h>
43 #include <linux/backing-dev.h>
44 #include <linux/pagevec.h>
45 #include <linux/cleancache.h>
46 
47 #include "ext4.h"
48 
49 /*
50  * Call ext4_decrypt on every single page, reusing the encryption
51  * context.
52  */
53 static void completion_pages(struct work_struct *work)
54 {
55 #ifdef CONFIG_EXT4_FS_ENCRYPTION
56 	struct ext4_crypto_ctx *ctx =
57 		container_of(work, struct ext4_crypto_ctx, work);
58 	struct bio	*bio	= ctx->bio;
59 	struct bio_vec	*bv;
60 	int		i;
61 
62 	bio_for_each_segment_all(bv, bio, i) {
63 		struct page *page = bv->bv_page;
64 
65 		int ret = ext4_decrypt(ctx, page);
66 		if (ret) {
67 			WARN_ON_ONCE(1);
68 			SetPageError(page);
69 		} else
70 			SetPageUptodate(page);
71 		unlock_page(page);
72 	}
73 	ext4_release_crypto_ctx(ctx);
74 	bio_put(bio);
75 #else
76 	BUG();
77 #endif
78 }
79 
80 static inline bool ext4_bio_encrypted(struct bio *bio)
81 {
82 #ifdef CONFIG_EXT4_FS_ENCRYPTION
83 	return unlikely(bio->bi_private != NULL);
84 #else
85 	return false;
86 #endif
87 }
88 
89 /*
90  * I/O completion handler for multipage BIOs.
91  *
92  * The mpage code never puts partial pages into a BIO (except for end-of-file).
93  * If a page does not map to a contiguous run of blocks then it simply falls
94  * back to block_read_full_page().
95  *
96  * Why is this?  If a page's completion depends on a number of different BIOs
97  * which can complete in any order (or at the same time) then determining the
98  * status of that page is hard.  See end_buffer_async_read() for the details.
99  * There is no point in duplicating all that complexity.
100  */
101 static void mpage_end_io(struct bio *bio, int err)
102 {
103 	struct bio_vec *bv;
104 	int i;
105 
106 	if (ext4_bio_encrypted(bio)) {
107 		struct ext4_crypto_ctx *ctx = bio->bi_private;
108 
109 		if (err) {
110 			ext4_release_crypto_ctx(ctx);
111 		} else {
112 			INIT_WORK(&ctx->work, completion_pages);
113 			ctx->bio = bio;
114 			queue_work(ext4_read_workqueue, &ctx->work);
115 			return;
116 		}
117 	}
118 	bio_for_each_segment_all(bv, bio, i) {
119 		struct page *page = bv->bv_page;
120 
121 		if (!err) {
122 			SetPageUptodate(page);
123 		} else {
124 			ClearPageUptodate(page);
125 			SetPageError(page);
126 		}
127 		unlock_page(page);
128 	}
129 
130 	bio_put(bio);
131 }
132 
133 int ext4_mpage_readpages(struct address_space *mapping,
134 			 struct list_head *pages, struct page *page,
135 			 unsigned nr_pages)
136 {
137 	struct bio *bio = NULL;
138 	unsigned page_idx;
139 	sector_t last_block_in_bio = 0;
140 
141 	struct inode *inode = mapping->host;
142 	const unsigned blkbits = inode->i_blkbits;
143 	const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
144 	const unsigned blocksize = 1 << blkbits;
145 	sector_t block_in_file;
146 	sector_t last_block;
147 	sector_t last_block_in_file;
148 	sector_t blocks[MAX_BUF_PER_PAGE];
149 	unsigned page_block;
150 	struct block_device *bdev = inode->i_sb->s_bdev;
151 	int length;
152 	unsigned relative_block = 0;
153 	struct ext4_map_blocks map;
154 
155 	map.m_pblk = 0;
156 	map.m_lblk = 0;
157 	map.m_len = 0;
158 	map.m_flags = 0;
159 
160 	for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
161 		int fully_mapped = 1;
162 		unsigned first_hole = blocks_per_page;
163 
164 		prefetchw(&page->flags);
165 		if (pages) {
166 			page = list_entry(pages->prev, struct page, lru);
167 			list_del(&page->lru);
168 			if (add_to_page_cache_lru(page, mapping,
169 						  page->index, GFP_KERNEL))
170 				goto next_page;
171 		}
172 
173 		if (page_has_buffers(page))
174 			goto confused;
175 
176 		block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
177 		last_block = block_in_file + nr_pages * blocks_per_page;
178 		last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
179 		if (last_block > last_block_in_file)
180 			last_block = last_block_in_file;
181 		page_block = 0;
182 
183 		/*
184 		 * Map blocks using the previous result first.
185 		 */
186 		if ((map.m_flags & EXT4_MAP_MAPPED) &&
187 		    block_in_file > map.m_lblk &&
188 		    block_in_file < (map.m_lblk + map.m_len)) {
189 			unsigned map_offset = block_in_file - map.m_lblk;
190 			unsigned last = map.m_len - map_offset;
191 
192 			for (relative_block = 0; ; relative_block++) {
193 				if (relative_block == last) {
194 					/* needed? */
195 					map.m_flags &= ~EXT4_MAP_MAPPED;
196 					break;
197 				}
198 				if (page_block == blocks_per_page)
199 					break;
200 				blocks[page_block] = map.m_pblk + map_offset +
201 					relative_block;
202 				page_block++;
203 				block_in_file++;
204 			}
205 		}
206 
207 		/*
208 		 * Then do more ext4_map_blocks() calls until we are
209 		 * done with this page.
210 		 */
211 		while (page_block < blocks_per_page) {
212 			if (block_in_file < last_block) {
213 				map.m_lblk = block_in_file;
214 				map.m_len = last_block - block_in_file;
215 
216 				if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
217 				set_error_page:
218 					SetPageError(page);
219 					zero_user_segment(page, 0,
220 							  PAGE_CACHE_SIZE);
221 					unlock_page(page);
222 					goto next_page;
223 				}
224 			}
225 			if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
226 				fully_mapped = 0;
227 				if (first_hole == blocks_per_page)
228 					first_hole = page_block;
229 				page_block++;
230 				block_in_file++;
231 				continue;
232 			}
233 			if (first_hole != blocks_per_page)
234 				goto confused;		/* hole -> non-hole */
235 
236 			/* Contiguous blocks? */
237 			if (page_block && blocks[page_block-1] != map.m_pblk-1)
238 				goto confused;
239 			for (relative_block = 0; ; relative_block++) {
240 				if (relative_block == map.m_len) {
241 					/* needed? */
242 					map.m_flags &= ~EXT4_MAP_MAPPED;
243 					break;
244 				} else if (page_block == blocks_per_page)
245 					break;
246 				blocks[page_block] = map.m_pblk+relative_block;
247 				page_block++;
248 				block_in_file++;
249 			}
250 		}
251 		if (first_hole != blocks_per_page) {
252 			zero_user_segment(page, first_hole << blkbits,
253 					  PAGE_CACHE_SIZE);
254 			if (first_hole == 0) {
255 				SetPageUptodate(page);
256 				unlock_page(page);
257 				goto next_page;
258 			}
259 		} else if (fully_mapped) {
260 			SetPageMappedToDisk(page);
261 		}
262 		if (fully_mapped && blocks_per_page == 1 &&
263 		    !PageUptodate(page) && cleancache_get_page(page) == 0) {
264 			SetPageUptodate(page);
265 			goto confused;
266 		}
267 
268 		/*
269 		 * This page will go to BIO.  Do we need to send this
270 		 * BIO off first?
271 		 */
272 		if (bio && (last_block_in_bio != blocks[0] - 1)) {
273 		submit_and_realloc:
274 			submit_bio(READ, bio);
275 			bio = NULL;
276 		}
277 		if (bio == NULL) {
278 			struct ext4_crypto_ctx *ctx = NULL;
279 
280 			if (ext4_encrypted_inode(inode) &&
281 			    S_ISREG(inode->i_mode)) {
282 				ctx = ext4_get_crypto_ctx(inode);
283 				if (IS_ERR(ctx))
284 					goto set_error_page;
285 			}
286 			bio = bio_alloc(GFP_KERNEL,
287 				min_t(int, nr_pages, bio_get_nr_vecs(bdev)));
288 			if (!bio) {
289 				if (ctx)
290 					ext4_release_crypto_ctx(ctx);
291 				goto set_error_page;
292 			}
293 			bio->bi_bdev = bdev;
294 			bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
295 			bio->bi_end_io = mpage_end_io;
296 			bio->bi_private = ctx;
297 		}
298 
299 		length = first_hole << blkbits;
300 		if (bio_add_page(bio, page, length, 0) < length)
301 			goto submit_and_realloc;
302 
303 		if (((map.m_flags & EXT4_MAP_BOUNDARY) &&
304 		     (relative_block == map.m_len)) ||
305 		    (first_hole != blocks_per_page)) {
306 			submit_bio(READ, bio);
307 			bio = NULL;
308 		} else
309 			last_block_in_bio = blocks[blocks_per_page - 1];
310 		goto next_page;
311 	confused:
312 		if (bio) {
313 			submit_bio(READ, bio);
314 			bio = NULL;
315 		}
316 		if (!PageUptodate(page))
317 			block_read_full_page(page, ext4_get_block);
318 		else
319 			unlock_page(page);
320 	next_page:
321 		if (pages)
322 			page_cache_release(page);
323 	}
324 	BUG_ON(pages && !list_empty(pages));
325 	if (bio)
326 		submit_bio(READ, bio);
327 	return 0;
328 }
329