1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/ext4/readpage.c 4 * 5 * Copyright (C) 2002, Linus Torvalds. 6 * Copyright (C) 2015, Google, Inc. 7 * 8 * This was originally taken from fs/mpage.c 9 * 10 * The intent is the ext4_mpage_readpages() function here is intended 11 * to replace mpage_readpages() in the general case, not just for 12 * encrypted files. It has some limitations (see below), where it 13 * will fall back to read_block_full_page(), but these limitations 14 * should only be hit when page_size != block_size. 15 * 16 * This will allow us to attach a callback function to support ext4 17 * encryption. 18 * 19 * If anything unusual happens, such as: 20 * 21 * - encountering a page which has buffers 22 * - encountering a page which has a non-hole after a hole 23 * - encountering a page with non-contiguous blocks 24 * 25 * then this code just gives up and calls the buffer_head-based read function. 26 * It does handle a page which has holes at the end - that is a common case: 27 * the end-of-file on blocksize < PAGE_SIZE setups. 28 * 29 */ 30 31 #include <linux/kernel.h> 32 #include <linux/export.h> 33 #include <linux/mm.h> 34 #include <linux/kdev_t.h> 35 #include <linux/gfp.h> 36 #include <linux/bio.h> 37 #include <linux/fs.h> 38 #include <linux/buffer_head.h> 39 #include <linux/blkdev.h> 40 #include <linux/highmem.h> 41 #include <linux/prefetch.h> 42 #include <linux/mpage.h> 43 #include <linux/writeback.h> 44 #include <linux/backing-dev.h> 45 #include <linux/pagevec.h> 46 #include <linux/cleancache.h> 47 48 #include "ext4.h" 49 50 static inline bool ext4_bio_encrypted(struct bio *bio) 51 { 52 #ifdef CONFIG_EXT4_FS_ENCRYPTION 53 return unlikely(bio->bi_private != NULL); 54 #else 55 return false; 56 #endif 57 } 58 59 /* 60 * I/O completion handler for multipage BIOs. 61 * 62 * The mpage code never puts partial pages into a BIO (except for end-of-file). 63 * If a page does not map to a contiguous run of blocks then it simply falls 64 * back to block_read_full_page(). 65 * 66 * Why is this? If a page's completion depends on a number of different BIOs 67 * which can complete in any order (or at the same time) then determining the 68 * status of that page is hard. See end_buffer_async_read() for the details. 69 * There is no point in duplicating all that complexity. 70 */ 71 static void mpage_end_io(struct bio *bio) 72 { 73 struct bio_vec *bv; 74 int i; 75 76 if (ext4_bio_encrypted(bio)) { 77 if (bio->bi_status) { 78 fscrypt_release_ctx(bio->bi_private); 79 } else { 80 fscrypt_enqueue_decrypt_bio(bio->bi_private, bio); 81 return; 82 } 83 } 84 bio_for_each_segment_all(bv, bio, i) { 85 struct page *page = bv->bv_page; 86 87 if (!bio->bi_status) { 88 SetPageUptodate(page); 89 } else { 90 ClearPageUptodate(page); 91 SetPageError(page); 92 } 93 unlock_page(page); 94 } 95 96 bio_put(bio); 97 } 98 99 int ext4_mpage_readpages(struct address_space *mapping, 100 struct list_head *pages, struct page *page, 101 unsigned nr_pages, bool is_readahead) 102 { 103 struct bio *bio = NULL; 104 sector_t last_block_in_bio = 0; 105 106 struct inode *inode = mapping->host; 107 const unsigned blkbits = inode->i_blkbits; 108 const unsigned blocks_per_page = PAGE_SIZE >> blkbits; 109 const unsigned blocksize = 1 << blkbits; 110 sector_t block_in_file; 111 sector_t last_block; 112 sector_t last_block_in_file; 113 sector_t blocks[MAX_BUF_PER_PAGE]; 114 unsigned page_block; 115 struct block_device *bdev = inode->i_sb->s_bdev; 116 int length; 117 unsigned relative_block = 0; 118 struct ext4_map_blocks map; 119 120 map.m_pblk = 0; 121 map.m_lblk = 0; 122 map.m_len = 0; 123 map.m_flags = 0; 124 125 for (; nr_pages; nr_pages--) { 126 int fully_mapped = 1; 127 unsigned first_hole = blocks_per_page; 128 129 prefetchw(&page->flags); 130 if (pages) { 131 page = lru_to_page(pages); 132 list_del(&page->lru); 133 if (add_to_page_cache_lru(page, mapping, page->index, 134 readahead_gfp_mask(mapping))) 135 goto next_page; 136 } 137 138 if (page_has_buffers(page)) 139 goto confused; 140 141 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); 142 last_block = block_in_file + nr_pages * blocks_per_page; 143 last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits; 144 if (last_block > last_block_in_file) 145 last_block = last_block_in_file; 146 page_block = 0; 147 148 /* 149 * Map blocks using the previous result first. 150 */ 151 if ((map.m_flags & EXT4_MAP_MAPPED) && 152 block_in_file > map.m_lblk && 153 block_in_file < (map.m_lblk + map.m_len)) { 154 unsigned map_offset = block_in_file - map.m_lblk; 155 unsigned last = map.m_len - map_offset; 156 157 for (relative_block = 0; ; relative_block++) { 158 if (relative_block == last) { 159 /* needed? */ 160 map.m_flags &= ~EXT4_MAP_MAPPED; 161 break; 162 } 163 if (page_block == blocks_per_page) 164 break; 165 blocks[page_block] = map.m_pblk + map_offset + 166 relative_block; 167 page_block++; 168 block_in_file++; 169 } 170 } 171 172 /* 173 * Then do more ext4_map_blocks() calls until we are 174 * done with this page. 175 */ 176 while (page_block < blocks_per_page) { 177 if (block_in_file < last_block) { 178 map.m_lblk = block_in_file; 179 map.m_len = last_block - block_in_file; 180 181 if (ext4_map_blocks(NULL, inode, &map, 0) < 0) { 182 set_error_page: 183 SetPageError(page); 184 zero_user_segment(page, 0, 185 PAGE_SIZE); 186 unlock_page(page); 187 goto next_page; 188 } 189 } 190 if ((map.m_flags & EXT4_MAP_MAPPED) == 0) { 191 fully_mapped = 0; 192 if (first_hole == blocks_per_page) 193 first_hole = page_block; 194 page_block++; 195 block_in_file++; 196 continue; 197 } 198 if (first_hole != blocks_per_page) 199 goto confused; /* hole -> non-hole */ 200 201 /* Contiguous blocks? */ 202 if (page_block && blocks[page_block-1] != map.m_pblk-1) 203 goto confused; 204 for (relative_block = 0; ; relative_block++) { 205 if (relative_block == map.m_len) { 206 /* needed? */ 207 map.m_flags &= ~EXT4_MAP_MAPPED; 208 break; 209 } else if (page_block == blocks_per_page) 210 break; 211 blocks[page_block] = map.m_pblk+relative_block; 212 page_block++; 213 block_in_file++; 214 } 215 } 216 if (first_hole != blocks_per_page) { 217 zero_user_segment(page, first_hole << blkbits, 218 PAGE_SIZE); 219 if (first_hole == 0) { 220 SetPageUptodate(page); 221 unlock_page(page); 222 goto next_page; 223 } 224 } else if (fully_mapped) { 225 SetPageMappedToDisk(page); 226 } 227 if (fully_mapped && blocks_per_page == 1 && 228 !PageUptodate(page) && cleancache_get_page(page) == 0) { 229 SetPageUptodate(page); 230 goto confused; 231 } 232 233 /* 234 * This page will go to BIO. Do we need to send this 235 * BIO off first? 236 */ 237 if (bio && (last_block_in_bio != blocks[0] - 1)) { 238 submit_and_realloc: 239 submit_bio(bio); 240 bio = NULL; 241 } 242 if (bio == NULL) { 243 struct fscrypt_ctx *ctx = NULL; 244 245 if (ext4_encrypted_inode(inode) && 246 S_ISREG(inode->i_mode)) { 247 ctx = fscrypt_get_ctx(inode, GFP_NOFS); 248 if (IS_ERR(ctx)) 249 goto set_error_page; 250 } 251 bio = bio_alloc(GFP_KERNEL, 252 min_t(int, nr_pages, BIO_MAX_PAGES)); 253 if (!bio) { 254 if (ctx) 255 fscrypt_release_ctx(ctx); 256 goto set_error_page; 257 } 258 bio_set_dev(bio, bdev); 259 bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9); 260 bio->bi_end_io = mpage_end_io; 261 bio->bi_private = ctx; 262 bio_set_op_attrs(bio, REQ_OP_READ, 263 is_readahead ? REQ_RAHEAD : 0); 264 } 265 266 length = first_hole << blkbits; 267 if (bio_add_page(bio, page, length, 0) < length) 268 goto submit_and_realloc; 269 270 if (((map.m_flags & EXT4_MAP_BOUNDARY) && 271 (relative_block == map.m_len)) || 272 (first_hole != blocks_per_page)) { 273 submit_bio(bio); 274 bio = NULL; 275 } else 276 last_block_in_bio = blocks[blocks_per_page - 1]; 277 goto next_page; 278 confused: 279 if (bio) { 280 submit_bio(bio); 281 bio = NULL; 282 } 283 if (!PageUptodate(page)) 284 block_read_full_page(page, ext4_get_block); 285 else 286 unlock_page(page); 287 next_page: 288 if (pages) 289 put_page(page); 290 } 291 BUG_ON(pages && !list_empty(pages)); 292 if (bio) 293 submit_bio(bio); 294 return 0; 295 } 296