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 #include <trace/events/android_fs.h>
49
50 /*
51 * Call ext4_decrypt on every single page, reusing the encryption
52 * context.
53 */
completion_pages(struct work_struct * work)54 static void completion_pages(struct work_struct *work)
55 {
56 #ifdef CONFIG_EXT4_FS_ENCRYPTION
57 struct ext4_crypto_ctx *ctx =
58 container_of(work, struct ext4_crypto_ctx, r.work);
59 struct bio *bio = ctx->r.bio;
60 struct bio_vec *bv;
61 int i;
62
63 bio_for_each_segment_all(bv, bio, i) {
64 struct page *page = bv->bv_page;
65
66 int ret = ext4_decrypt(page);
67 if (ret) {
68 WARN_ON_ONCE(1);
69 SetPageError(page);
70 } else
71 SetPageUptodate(page);
72 unlock_page(page);
73 }
74 ext4_release_crypto_ctx(ctx);
75 bio_put(bio);
76 #else
77 BUG();
78 #endif
79 }
80
ext4_bio_encrypted(struct bio * bio)81 static inline bool ext4_bio_encrypted(struct bio *bio)
82 {
83 #ifdef CONFIG_EXT4_FS_ENCRYPTION
84 return unlikely(bio->bi_private != NULL);
85 #else
86 return false;
87 #endif
88 }
89
90 static void
ext4_trace_read_completion(struct bio * bio)91 ext4_trace_read_completion(struct bio *bio)
92 {
93 struct page *first_page = bio->bi_io_vec[0].bv_page;
94
95 if (first_page != NULL)
96 trace_android_fs_dataread_end(first_page->mapping->host,
97 page_offset(first_page),
98 bio->bi_iter.bi_size);
99 }
100
101 /*
102 * I/O completion handler for multipage BIOs.
103 *
104 * The mpage code never puts partial pages into a BIO (except for end-of-file).
105 * If a page does not map to a contiguous run of blocks then it simply falls
106 * back to block_read_full_page().
107 *
108 * Why is this? If a page's completion depends on a number of different BIOs
109 * which can complete in any order (or at the same time) then determining the
110 * status of that page is hard. See end_buffer_async_read() for the details.
111 * There is no point in duplicating all that complexity.
112 */
mpage_end_io(struct bio * bio)113 static void mpage_end_io(struct bio *bio)
114 {
115 struct bio_vec *bv;
116 int i;
117
118 if (trace_android_fs_dataread_start_enabled())
119 ext4_trace_read_completion(bio);
120
121 if (ext4_bio_encrypted(bio)) {
122 struct ext4_crypto_ctx *ctx = bio->bi_private;
123
124 if (bio->bi_error) {
125 ext4_release_crypto_ctx(ctx);
126 } else {
127 INIT_WORK(&ctx->r.work, completion_pages);
128 ctx->r.bio = bio;
129 queue_work(ext4_read_workqueue, &ctx->r.work);
130 return;
131 }
132 }
133 bio_for_each_segment_all(bv, bio, i) {
134 struct page *page = bv->bv_page;
135
136 if (!bio->bi_error) {
137 SetPageUptodate(page);
138 } else {
139 ClearPageUptodate(page);
140 SetPageError(page);
141 }
142 unlock_page(page);
143 }
144
145 bio_put(bio);
146 }
147
148 static void
ext4_submit_bio_read(struct bio * bio)149 ext4_submit_bio_read(struct bio *bio)
150 {
151 if (trace_android_fs_dataread_start_enabled()) {
152 struct page *first_page = bio->bi_io_vec[0].bv_page;
153
154 if (first_page != NULL) {
155 char *path, pathbuf[MAX_TRACE_PATHBUF_LEN];
156
157 path = android_fstrace_get_pathname(pathbuf,
158 MAX_TRACE_PATHBUF_LEN,
159 first_page->mapping->host);
160 trace_android_fs_dataread_start(
161 first_page->mapping->host,
162 page_offset(first_page),
163 bio->bi_iter.bi_size,
164 current->pid,
165 path,
166 current->comm);
167 }
168 }
169 submit_bio(READ, bio);
170 }
171
ext4_mpage_readpages(struct address_space * mapping,struct list_head * pages,struct page * page,unsigned nr_pages)172 int ext4_mpage_readpages(struct address_space *mapping,
173 struct list_head *pages, struct page *page,
174 unsigned nr_pages)
175 {
176 struct bio *bio = NULL;
177 unsigned page_idx;
178 sector_t last_block_in_bio = 0;
179
180 struct inode *inode = mapping->host;
181 const unsigned blkbits = inode->i_blkbits;
182 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
183 const unsigned blocksize = 1 << blkbits;
184 sector_t block_in_file;
185 sector_t last_block;
186 sector_t last_block_in_file;
187 sector_t blocks[MAX_BUF_PER_PAGE];
188 unsigned page_block;
189 struct block_device *bdev = inode->i_sb->s_bdev;
190 int length;
191 unsigned relative_block = 0;
192 struct ext4_map_blocks map;
193
194 map.m_pblk = 0;
195 map.m_lblk = 0;
196 map.m_len = 0;
197 map.m_flags = 0;
198
199 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
200 int fully_mapped = 1;
201 unsigned first_hole = blocks_per_page;
202
203 prefetchw(&page->flags);
204 if (pages) {
205 page = list_entry(pages->prev, struct page, lru);
206 list_del(&page->lru);
207 if (add_to_page_cache_lru(page, mapping, page->index,
208 mapping_gfp_constraint(mapping, GFP_KERNEL)))
209 goto next_page;
210 }
211
212 if (page_has_buffers(page))
213 goto confused;
214
215 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
216 last_block = block_in_file + nr_pages * blocks_per_page;
217 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
218 if (last_block > last_block_in_file)
219 last_block = last_block_in_file;
220 page_block = 0;
221
222 /*
223 * Map blocks using the previous result first.
224 */
225 if ((map.m_flags & EXT4_MAP_MAPPED) &&
226 block_in_file > map.m_lblk &&
227 block_in_file < (map.m_lblk + map.m_len)) {
228 unsigned map_offset = block_in_file - map.m_lblk;
229 unsigned last = map.m_len - map_offset;
230
231 for (relative_block = 0; ; relative_block++) {
232 if (relative_block == last) {
233 /* needed? */
234 map.m_flags &= ~EXT4_MAP_MAPPED;
235 break;
236 }
237 if (page_block == blocks_per_page)
238 break;
239 blocks[page_block] = map.m_pblk + map_offset +
240 relative_block;
241 page_block++;
242 block_in_file++;
243 }
244 }
245
246 /*
247 * Then do more ext4_map_blocks() calls until we are
248 * done with this page.
249 */
250 while (page_block < blocks_per_page) {
251 if (block_in_file < last_block) {
252 map.m_lblk = block_in_file;
253 map.m_len = last_block - block_in_file;
254
255 if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
256 set_error_page:
257 SetPageError(page);
258 zero_user_segment(page, 0,
259 PAGE_CACHE_SIZE);
260 unlock_page(page);
261 goto next_page;
262 }
263 }
264 if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
265 fully_mapped = 0;
266 if (first_hole == blocks_per_page)
267 first_hole = page_block;
268 page_block++;
269 block_in_file++;
270 continue;
271 }
272 if (first_hole != blocks_per_page)
273 goto confused; /* hole -> non-hole */
274
275 /* Contiguous blocks? */
276 if (page_block && blocks[page_block-1] != map.m_pblk-1)
277 goto confused;
278 for (relative_block = 0; ; relative_block++) {
279 if (relative_block == map.m_len) {
280 /* needed? */
281 map.m_flags &= ~EXT4_MAP_MAPPED;
282 break;
283 } else if (page_block == blocks_per_page)
284 break;
285 blocks[page_block] = map.m_pblk+relative_block;
286 page_block++;
287 block_in_file++;
288 }
289 }
290 if (first_hole != blocks_per_page) {
291 zero_user_segment(page, first_hole << blkbits,
292 PAGE_CACHE_SIZE);
293 if (first_hole == 0) {
294 SetPageUptodate(page);
295 unlock_page(page);
296 goto next_page;
297 }
298 } else if (fully_mapped) {
299 SetPageMappedToDisk(page);
300 }
301 if (fully_mapped && blocks_per_page == 1 &&
302 !PageUptodate(page) && cleancache_get_page(page) == 0) {
303 SetPageUptodate(page);
304 goto confused;
305 }
306
307 /*
308 * This page will go to BIO. Do we need to send this
309 * BIO off first?
310 */
311 if (bio && (last_block_in_bio != blocks[0] - 1)) {
312 submit_and_realloc:
313 ext4_submit_bio_read(bio);
314 bio = NULL;
315 }
316 if (bio == NULL) {
317 struct ext4_crypto_ctx *ctx = NULL;
318
319 if (ext4_encrypted_inode(inode) &&
320 S_ISREG(inode->i_mode)) {
321 ctx = ext4_get_crypto_ctx(inode, GFP_NOFS);
322 if (IS_ERR(ctx))
323 goto set_error_page;
324 }
325 bio = bio_alloc(GFP_KERNEL,
326 min_t(int, nr_pages, BIO_MAX_PAGES));
327 if (!bio) {
328 if (ctx)
329 ext4_release_crypto_ctx(ctx);
330 goto set_error_page;
331 }
332 bio->bi_bdev = bdev;
333 bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
334 bio->bi_end_io = mpage_end_io;
335 bio->bi_private = ctx;
336 }
337
338 length = first_hole << blkbits;
339 if (bio_add_page(bio, page, length, 0) < length)
340 goto submit_and_realloc;
341
342 if (((map.m_flags & EXT4_MAP_BOUNDARY) &&
343 (relative_block == map.m_len)) ||
344 (first_hole != blocks_per_page)) {
345 ext4_submit_bio_read(bio);
346 bio = NULL;
347 } else
348 last_block_in_bio = blocks[blocks_per_page - 1];
349 goto next_page;
350 confused:
351 if (bio) {
352 ext4_submit_bio_read(bio);
353 bio = NULL;
354 }
355 if (!PageUptodate(page))
356 block_read_full_page(page, ext4_get_block);
357 else
358 unlock_page(page);
359 next_page:
360 if (pages)
361 page_cache_release(page);
362 }
363 BUG_ON(pages && !list_empty(pages));
364 if (bio)
365 ext4_submit_bio_read(bio);
366 return 0;
367 }
368