fs/ext4/readpage.c

   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  */
  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
  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
  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  */
 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
 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                         trace_android_fs_dataread_start(
 156                                 first_page->mapping->host,
 157                                 page_offset(first_page),
 158                                 bio->bi_iter.bi_size,
 159                                 current->pid,
 160                                 current->comm);
 161                 }
 162         }
 163         submit_bio(READ, bio);
 164 }
 165
 166 int ext4_mpage_readpages(struct address_space *mapping,
 167                          struct list_head *pages, struct page *page,
 168                          unsigned nr_pages)
 169 {
 170         struct bio *bio = NULL;
 171         unsigned page_idx;
 172         sector_t last_block_in_bio = 0;
 173
 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;
 179         sector_t last_block;
 180         sector_t last_block_in_file;
 181         sector_t blocks[MAX_BUF_PER_PAGE];
 182         unsigned page_block;
 183         struct block_device *bdev = inode->i_sb->s_bdev;
 184         int length;
 185         unsigned relative_block = 0;
 186         struct ext4_map_blocks map;
 187
 188         map.m_pblk = 0;
 189         map.m_lblk = 0;
 190         map.m_len = 0;
 191         map.m_flags = 0;
 192
 193         for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
 194                 int fully_mapped = 1;
 195                 unsigned first_hole = blocks_per_page;
 196
 197                 prefetchw(&page->flags);
 198                 if (pages) {
 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)))
 203                                 goto next_page;
 204                 }
 205
 206                 if (page_has_buffers(page))
 207                         goto confused;
 208
 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;
 214                 page_block = 0;
 215
 216                 /*
 217                  * Map blocks using the previous result first.
 218                  */
 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;
 224
 225                         for (relative_block = 0; ; relative_block++) {
 226                                 if (relative_block == last) {
 227                                         /* needed? */
 228                                         map.m_flags &= ~EXT4_MAP_MAPPED;
 229                                         break;
 230                                 }
 231                                 if (page_block == blocks_per_page)
 232                                         break;
 233                                 blocks[page_block] = map.m_pblk + map_offset +
 234                                         relative_block;
 235                                 page_block++;
 236                                 block_in_file++;
 237                         }
 238                 }
 239
 240                 /*
 241                  * Then do more ext4_map_blocks() calls until we are
 242                  * done with this page.
 243                  */
 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;
 248
 249                                 if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
 250                                 set_error_page:
 251                                         SetPageError(page);
 252                                         zero_user_segment(page, 0,
 253                                                           PAGE_CACHE_SIZE);
 254                                         unlock_page(page);
 255                                         goto next_page;
 256                                 }
 257                         }
 258                         if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
 259                                 fully_mapped = 0;
 260                                 if (first_hole == blocks_per_page)
 261                                         first_hole = page_block;
 262                                 page_block++;
 263                                 block_in_file++;
 264                                 continue;
 265                         }
 266                         if (first_hole != blocks_per_page)
 267                                 goto confused;          /* hole -> non-hole */
 268
 269                         /* Contiguous blocks? */
 270                         if (page_block && blocks[page_block-1] != map.m_pblk-1)
 271                                 goto confused;
 272                         for (relative_block = 0; ; relative_block++) {
 273                                 if (relative_block == map.m_len) {
 274                                         /* needed? */
 275                                         map.m_flags &= ~EXT4_MAP_MAPPED;
 276                                         break;
 277                                 } else if (page_block == blocks_per_page)
 278                                         break;
 279                                 blocks[page_block] = map.m_pblk+relative_block;
 280                                 page_block++;
 281                                 block_in_file++;
 282                         }
 283                 }
 284                 if (first_hole != blocks_per_page) {
 285                         zero_user_segment(page, first_hole << blkbits,
 286                                           PAGE_CACHE_SIZE);
 287                         if (first_hole == 0) {
 288                                 SetPageUptodate(page);
 289                                 unlock_page(page);
 290                                 goto next_page;
 291                         }
 292                 } else if (fully_mapped) {
 293                         SetPageMappedToDisk(page);
 294                 }
 295                 if (fully_mapped && blocks_per_page == 1 &&
 296                     !PageUptodate(page) && cleancache_get_page(page) == 0) {
 297                         SetPageUptodate(page);
 298                         goto confused;
 299                 }
 300
 301                 /*
 302                  * This page will go to BIO.  Do we need to send this
 303                  * BIO off first?
 304                  */
 305                 if (bio && (last_block_in_bio != blocks[0] - 1)) {
 306                 submit_and_realloc:
 307                         ext4_submit_bio_read(bio);
 308                         bio = NULL;
 309                 }
 310                 if (bio == NULL) {
 311                         struct ext4_crypto_ctx *ctx = NULL;
 312
 313                         if (ext4_encrypted_inode(inode) &&
 314                             S_ISREG(inode->i_mode)) {
 315                                 ctx = ext4_get_crypto_ctx(inode);
 316                                 if (IS_ERR(ctx))
 317                                         goto set_error_page;
 318                         }
 319                         bio = bio_alloc(GFP_KERNEL,
 320                                 min_t(int, nr_pages, BIO_MAX_PAGES));
 321                         if (!bio) {
 322                                 if (ctx)
 323                                         ext4_release_crypto_ctx(ctx);
 324                                 goto set_error_page;
 325                         }
 326                         bio->bi_bdev = bdev;
 327                         bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
 328                         bio->bi_end_io = mpage_end_io;
 329                         bio->bi_private = ctx;
 330                 }
 331
 332                 length = first_hole << blkbits;
 333                 if (bio_add_page(bio, page, length, 0) < length)
 334                         goto submit_and_realloc;
 335
 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);
 340                         bio = NULL;
 341                 } else
 342                         last_block_in_bio = blocks[blocks_per_page - 1];
 343                 goto next_page;
 344         confused:
 345                 if (bio) {
 346                         ext4_submit_bio_read(bio);
 347                         bio = NULL;
 348                 }
 349                 if (!PageUptodate(page))
 350                         block_read_full_page(page, ext4_get_block);
 351                 else
 352                         unlock_page(page);
 353         next_page:
 354                 if (pages)
 355                         page_cache_release(page);
 356         }
 357         BUG_ON(pages && !list_empty(pages));
 358         if (bio)
 359                 ext4_submit_bio_read(bio);
 360         return 0;
 361 }