ext4 crypto: add ext4_mpage_readpages()
authorTheodore Ts'o <tytso@mit.edu>
Wed, 8 Apr 2015 04:00:32 +0000 (00:00 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Wed, 8 Apr 2015 04:00:32 +0000 (00:00 -0400)
This takes code from fs/mpage.c and optimizes it for ext4.  Its
primary reason is to allow us to more easily add encryption to ext4's
read path in an efficient manner.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
fs/ext4/Makefile
fs/ext4/ext4.h
fs/ext4/inode.c
fs/ext4/readpage.c [new file with mode: 0644]

index 0310fec2ee3dbd39ddf3b25ae47b89147fe40031..cd6f50fce27817d38f46f7c865d2a73f1e228165 100644 (file)
@@ -8,7 +8,7 @@ ext4-y  := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \
                ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
                ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \
                mmp.o indirect.o extents_status.o xattr.o xattr_user.o \
-               xattr_trusted.o inline.o
+               xattr_trusted.o inline.o readpage.o
 
 ext4-$(CONFIG_EXT4_FS_POSIX_ACL)       += acl.o
 ext4-$(CONFIG_EXT4_FS_SECURITY)                += xattr_security.o
index f63c3d5805c4c156ad3ed412cbecf85e700cf9d2..f7f3f5871796d9c44ec689a74518ae46eeb77215 100644 (file)
@@ -2699,6 +2699,10 @@ static inline void ext4_set_de_type(struct super_block *sb,
                de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
 }
 
+/* readpages.c */
+extern int ext4_mpage_readpages(struct address_space *mapping,
+                               struct list_head *pages, struct page *page,
+                               unsigned nr_pages);
 
 /* symlink.c */
 extern const struct inode_operations ext4_symlink_inode_operations;
index ff0c111e52eba044c5a94ff26b1d543eaa86bbe9..cd3009152ae28de43f0d3a80b7a7235a417b77d5 100644 (file)
@@ -2820,7 +2820,7 @@ static int ext4_readpage(struct file *file, struct page *page)
                ret = ext4_readpage_inline(inode, page);
 
        if (ret == -EAGAIN)
-               return mpage_readpage(page, ext4_get_block);
+               return ext4_mpage_readpages(page->mapping, NULL, page, 1);
 
        return ret;
 }
@@ -2835,7 +2835,7 @@ ext4_readpages(struct file *file, struct address_space *mapping,
        if (ext4_has_inline_data(inode))
                return 0;
 
-       return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
+       return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
 }
 
 static void ext4_invalidatepage(struct page *page, unsigned int offset,
diff --git a/fs/ext4/readpage.c b/fs/ext4/readpage.c
new file mode 100644 (file)
index 0000000..fff9fe6
--- /dev/null
@@ -0,0 +1,264 @@
+/*
+ * linux/fs/ext4/readpage.c
+ *
+ * Copyright (C) 2002, Linus Torvalds.
+ * Copyright (C) 2015, Google, Inc.
+ *
+ * This was originally taken from fs/mpage.c
+ *
+ * The intent is the ext4_mpage_readpages() function here is intended
+ * to replace mpage_readpages() in the general case, not just for
+ * encrypted files.  It has some limitations (see below), where it
+ * will fall back to read_block_full_page(), but these limitations
+ * should only be hit when page_size != block_size.
+ *
+ * This will allow us to attach a callback function to support ext4
+ * encryption.
+ *
+ * If anything unusual happens, such as:
+ *
+ * - encountering a page which has buffers
+ * - encountering a page which has a non-hole after a hole
+ * - encountering a page with non-contiguous blocks
+ *
+ * then this code just gives up and calls the buffer_head-based read function.
+ * It does handle a page which has holes at the end - that is a common case:
+ * the end-of-file on blocksize < PAGE_CACHE_SIZE setups.
+ *
+ */
+
+#include <linux/kernel.h>
+#include <linux/export.h>
+#include <linux/mm.h>
+#include <linux/kdev_t.h>
+#include <linux/gfp.h>
+#include <linux/bio.h>
+#include <linux/fs.h>
+#include <linux/buffer_head.h>
+#include <linux/blkdev.h>
+#include <linux/highmem.h>
+#include <linux/prefetch.h>
+#include <linux/mpage.h>
+#include <linux/writeback.h>
+#include <linux/backing-dev.h>
+#include <linux/pagevec.h>
+#include <linux/cleancache.h>
+
+#include "ext4.h"
+
+/*
+ * I/O completion handler for multipage BIOs.
+ *
+ * The mpage code never puts partial pages into a BIO (except for end-of-file).
+ * If a page does not map to a contiguous run of blocks then it simply falls
+ * back to block_read_full_page().
+ *
+ * Why is this?  If a page's completion depends on a number of different BIOs
+ * which can complete in any order (or at the same time) then determining the
+ * status of that page is hard.  See end_buffer_async_read() for the details.
+ * There is no point in duplicating all that complexity.
+ */
+static void mpage_end_io(struct bio *bio, int err)
+{
+       struct bio_vec *bv;
+       int i;
+
+       bio_for_each_segment_all(bv, bio, i) {
+               struct page *page = bv->bv_page;
+
+               if (!err) {
+                       SetPageUptodate(page);
+               } else {
+                       ClearPageUptodate(page);
+                       SetPageError(page);
+               }
+               unlock_page(page);
+       }
+
+       bio_put(bio);
+}
+
+int ext4_mpage_readpages(struct address_space *mapping,
+                        struct list_head *pages, struct page *page,
+                        unsigned nr_pages)
+{
+       struct bio *bio = NULL;
+       unsigned page_idx;
+       sector_t last_block_in_bio = 0;
+
+       struct inode *inode = mapping->host;
+       const unsigned blkbits = inode->i_blkbits;
+       const unsigned blocks_per_page = PAGE_CACHE_SIZE >> blkbits;
+       const unsigned blocksize = 1 << blkbits;
+       sector_t block_in_file;
+       sector_t last_block;
+       sector_t last_block_in_file;
+       sector_t blocks[MAX_BUF_PER_PAGE];
+       unsigned page_block;
+       struct block_device *bdev = inode->i_sb->s_bdev;
+       int length;
+       unsigned relative_block = 0;
+       struct ext4_map_blocks map;
+
+       map.m_pblk = 0;
+       map.m_lblk = 0;
+       map.m_len = 0;
+       map.m_flags = 0;
+
+       for (page_idx = 0; nr_pages; page_idx++, nr_pages--) {
+               int fully_mapped = 1;
+               unsigned first_hole = blocks_per_page;
+
+               prefetchw(&page->flags);
+               if (pages) {
+                       page = list_entry(pages->prev, struct page, lru);
+                       list_del(&page->lru);
+                       if (add_to_page_cache_lru(page, mapping,
+                                                 page->index, GFP_KERNEL))
+                               goto next_page;
+               }
+
+               if (page_has_buffers(page))
+                       goto confused;
+
+               block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
+               last_block = block_in_file + nr_pages * blocks_per_page;
+               last_block_in_file = (i_size_read(inode) + blocksize - 1) >> blkbits;
+               if (last_block > last_block_in_file)
+                       last_block = last_block_in_file;
+               page_block = 0;
+
+               /*
+                * Map blocks using the previous result first.
+                */
+               if ((map.m_flags & EXT4_MAP_MAPPED) &&
+                   block_in_file > map.m_lblk &&
+                   block_in_file < (map.m_lblk + map.m_len)) {
+                       unsigned map_offset = block_in_file - map.m_lblk;
+                       unsigned last = map.m_len - map_offset;
+
+                       for (relative_block = 0; ; relative_block++) {
+                               if (relative_block == last) {
+                                       /* needed? */
+                                       map.m_flags &= ~EXT4_MAP_MAPPED;
+                                       break;
+                               }
+                               if (page_block == blocks_per_page)
+                                       break;
+                               blocks[page_block] = map.m_pblk + map_offset +
+                                       relative_block;
+                               page_block++;
+                               block_in_file++;
+                       }
+               }
+
+               /*
+                * Then do more ext4_map_blocks() calls until we are
+                * done with this page.
+                */
+               while (page_block < blocks_per_page) {
+                       if (block_in_file < last_block) {
+                               map.m_lblk = block_in_file;
+                               map.m_len = last_block - block_in_file;
+
+                               if (ext4_map_blocks(NULL, inode, &map, 0) < 0) {
+                               set_error_page:
+                                       SetPageError(page);
+                                       zero_user_segment(page, 0,
+                                                         PAGE_CACHE_SIZE);
+                                       unlock_page(page);
+                                       goto next_page;
+                               }
+                       }
+                       if ((map.m_flags & EXT4_MAP_MAPPED) == 0) {
+                               fully_mapped = 0;
+                               if (first_hole == blocks_per_page)
+                                       first_hole = page_block;
+                               page_block++;
+                               block_in_file++;
+                               continue;
+                       }
+                       if (first_hole != blocks_per_page)
+                               goto confused;          /* hole -> non-hole */
+
+                       /* Contiguous blocks? */
+                       if (page_block && blocks[page_block-1] != map.m_pblk-1)
+                               goto confused;
+                       for (relative_block = 0; ; relative_block++) {
+                               if (relative_block == map.m_len) {
+                                       /* needed? */
+                                       map.m_flags &= ~EXT4_MAP_MAPPED;
+                                       break;
+                               } else if (page_block == blocks_per_page)
+                                       break;
+                               blocks[page_block] = map.m_pblk+relative_block;
+                               page_block++;
+                               block_in_file++;
+                       }
+               }
+               if (first_hole != blocks_per_page) {
+                       zero_user_segment(page, first_hole << blkbits,
+                                         PAGE_CACHE_SIZE);
+                       if (first_hole == 0) {
+                               SetPageUptodate(page);
+                               unlock_page(page);
+                               goto next_page;
+                       }
+               } else if (fully_mapped) {
+                       SetPageMappedToDisk(page);
+               }
+               if (fully_mapped && blocks_per_page == 1 &&
+                   !PageUptodate(page) && cleancache_get_page(page) == 0) {
+                       SetPageUptodate(page);
+                       goto confused;
+               }
+
+               /*
+                * This page will go to BIO.  Do we need to send this
+                * BIO off first?
+                */
+               if (bio && (last_block_in_bio != blocks[0] - 1)) {
+               submit_and_realloc:
+                       submit_bio(READ, bio);
+                       bio = NULL;
+               }
+               if (bio == NULL) {
+                       bio = bio_alloc(GFP_KERNEL,
+                               min_t(int, nr_pages, bio_get_nr_vecs(bdev)));
+                       if (!bio)
+                               goto set_error_page;
+                       bio->bi_bdev = bdev;
+                       bio->bi_iter.bi_sector = blocks[0] << (blkbits - 9);
+                       bio->bi_end_io = mpage_end_io;
+               }
+
+               length = first_hole << blkbits;
+               if (bio_add_page(bio, page, length, 0) < length)
+                       goto submit_and_realloc;
+
+               if (((map.m_flags & EXT4_MAP_BOUNDARY) &&
+                    (relative_block == map.m_len)) ||
+                   (first_hole != blocks_per_page)) {
+                       submit_bio(READ, bio);
+                       bio = NULL;
+               } else
+                       last_block_in_bio = blocks[blocks_per_page - 1];
+               goto next_page;
+       confused:
+               if (bio) {
+                       submit_bio(READ, bio);
+                       bio = NULL;
+               }
+               if (!PageUptodate(page))
+                       block_read_full_page(page, ext4_get_block);
+               else
+                       unlock_page(page);
+       next_page:
+               if (pages)
+                       page_cache_release(page);
+       }
+       BUG_ON(pages && !list_empty(pages));
+       if (bio)
+               submit_bio(READ, bio);
+       return 0;
+}