2 * linux/fs/ext4/readpage.c
4 * Copyright (C) 2002, Linus Torvalds.
5 * Copyright (C) 2015, Google, Inc.
7 * This was originally taken from fs/mpage.c
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.
15 * This will allow us to attach a callback function to support ext4
18 * If anything unusual happens, such as:
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
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.
30 #include <linux/kernel.h>
31 #include <linux/export.h>
33 #include <linux/kdev_t.h>
34 #include <linux/gfp.h>
35 #include <linux/bio.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>
48 #include <trace/events/android_fs.h>
51 * Call ext4_decrypt on every single page, reusing the encryption
54 static void completion_pages(struct work_struct *work)
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;
63 bio_for_each_segment_all(bv, bio, i) {
64 struct page *page = bv->bv_page;
66 int ret = ext4_decrypt(page);
71 SetPageUptodate(page);
74 ext4_release_crypto_ctx(ctx);
81 static inline bool ext4_bio_encrypted(struct bio *bio)
83 #ifdef CONFIG_EXT4_FS_ENCRYPTION
84 return unlikely(bio->bi_private != NULL);
91 ext4_trace_read_completion(struct bio *bio)
93 struct page *first_page = bio->bi_io_vec[0].bv_page;
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);
102 * I/O completion handler for multipage BIOs.
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().
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.
113 static void mpage_end_io(struct bio *bio)
118 if (trace_android_fs_dataread_start_enabled())
119 ext4_trace_read_completion(bio);
121 if (ext4_bio_encrypted(bio)) {
122 struct ext4_crypto_ctx *ctx = bio->bi_private;
125 ext4_release_crypto_ctx(ctx);
127 INIT_WORK(&ctx->r.work, completion_pages);
129 queue_work(ext4_read_workqueue, &ctx->r.work);
133 bio_for_each_segment_all(bv, bio, i) {
134 struct page *page = bv->bv_page;
136 if (!bio->bi_error) {
137 SetPageUptodate(page);
139 ClearPageUptodate(page);
149 ext4_submit_bio_read(struct bio *bio)
151 if (trace_android_fs_dataread_start_enabled()) {
152 struct page *first_page = bio->bi_io_vec[0].bv_page;
154 if (first_page != NULL) {
155 trace_android_fs_dataread_start(
156 first_page->mapping->host,
157 page_offset(first_page),
158 bio->bi_iter.bi_size,
163 submit_bio(READ, bio);
166 int ext4_mpage_readpages(struct address_space *mapping,
167 struct list_head *pages, struct page *page,
170 struct bio *bio = NULL;
172 sector_t last_block_in_bio = 0;
174 struct inode *inode = mapping->host;
175 const unsigned blkbits = inode->i_blkbits;
176 const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
177 const unsigned blocksize = 1 << blkbits;
178 sector_t block_in_file;
180 sector_t last_block_in_file;
181 sector_t blocks[MAX_BUF_PER_PAGE];
183 struct block_device *bdev = inode->i_sb->s_bdev;
185 unsigned relative_block = 0;
186 struct ext4_map_blocks map;
193 for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
194 int fully_mapped = 1;
195 unsigned first_hole = blocks_per_page;
197 prefetchw(&page->flags);
199 page = list_entry(pages->prev, struct page, lru);
200 list_del(&page->lru);
201 if (add_to_page_cache_lru(page, mapping, page->index,
202 mapping_gfp_constraint(mapping, GFP_KERNEL)))
206 if (page_has_buffers(page))
209 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
210 last_block = block_in_file + nr_pages * blocks_per_page;
211 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
212 if (last_block > last_block_in_file)
213 last_block = last_block_in_file;
217 * Map blocks using the previous result first.
219 if ((map.m_flags & EXT4_MAP_MAPPED) &&
220 block_in_file > map.m_lblk &&
221 block_in_file < (map.m_lblk + map.m_len)) {
222 unsigned map_offset = block_in_file - map.m_lblk;
223 unsigned last = map.m_len - map_offset;
225 for (relative_block = 0; ; relative_block++) {
226 if (relative_block == last) {
228 map.m_flags &= ~EXT4_MAP_MAPPED;
231 if (page_block == blocks_per_page)
233 blocks[page_block] = map.m_pblk + map_offset +
241 * Then do more ext4_map_blocks() calls until we are
242 * done with this page.
244 while (page_block < blocks_per_page) {
245 if (block_in_file < last_block) {
246 map.m_lblk = block_in_file;
247 map.m_len = last_block - block_in_file;
249 if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
252 zero_user_segment(page, 0,
258 if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
260 if (first_hole == blocks_per_page)
261 first_hole = page_block;
266 if (first_hole != blocks_per_page)
267 goto confused; /* hole -> non-hole */
269 /* Contiguous blocks? */
270 if (page_block && blocks[page_block-1] != map.m_pblk-1)
272 for (relative_block = 0; ; relative_block++) {
273 if (relative_block == map.m_len) {
275 map.m_flags &= ~EXT4_MAP_MAPPED;
277 } else if (page_block == blocks_per_page)
279 blocks[page_block] = map.m_pblk+relative_block;
284 if (first_hole != blocks_per_page) {
285 zero_user_segment(page, first_hole << blkbits,
287 if (first_hole == 0) {
288 SetPageUptodate(page);
292 } else if (fully_mapped) {
293 SetPageMappedToDisk(page);
295 if (fully_mapped && blocks_per_page == 1 &&
296 !PageUptodate(page) && cleancache_get_page(page) == 0) {
297 SetPageUptodate(page);
302 * This page will go to BIO. Do we need to send this
305 if (bio && (last_block_in_bio != blocks[0] - 1)) {
307 ext4_submit_bio_read(bio);
311 struct ext4_crypto_ctx *ctx = NULL;
313 if (ext4_encrypted_inode(inode) &&
314 S_ISREG(inode->i_mode)) {
315 ctx = ext4_get_crypto_ctx(inode);
319 bio = bio_alloc(GFP_KERNEL,
320 min_t(int, nr_pages, BIO_MAX_PAGES));
323 ext4_release_crypto_ctx(ctx);
327 bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
328 bio->bi_end_io = mpage_end_io;
329 bio->bi_private = ctx;
332 length = first_hole << blkbits;
333 if (bio_add_page(bio, page, length, 0) < length)
334 goto submit_and_realloc;
336 if (((map.m_flags & EXT4_MAP_BOUNDARY) &&
337 (relative_block == map.m_len)) ||
338 (first_hole != blocks_per_page)) {
339 ext4_submit_bio_read(bio);
342 last_block_in_bio = blocks[blocks_per_page - 1];
346 ext4_submit_bio_read(bio);
349 if (!PageUptodate(page))
350 block_read_full_page(page, ext4_get_block);
355 page_cache_release(page);
357 BUG_ON(pages && !list_empty(pages));
359 ext4_submit_bio_read(bio);