erofs: convert all uncompressed cases to iomap
authorGao Xiang <hsiangkao@linux.alibaba.com>
Thu, 5 Aug 2021 00:36:01 +0000 (08:36 +0800)
committerGao Xiang <hsiangkao@linux.alibaba.com>
Mon, 9 Aug 2021 16:18:46 +0000 (00:18 +0800)
Since tail-packing inline has been supported by iomap now, let's
convert all EROFS uncompressed data I/O to iomap, which is pretty
straight-forward.

Link: https://lore.kernel.org/r/20210805003601.183063-4-hsiangkao@linux.alibaba.com
Cc: linux-fsdevel@vger.kernel.org
Reviewed-by: Chao Yu <chao@kernel.org>
Signed-off-by: Gao Xiang <hsiangkao@linux.alibaba.com>
fs/erofs/data.c

index fb2e554c861b77ab4c41aa8b501829ed59930d3b..b2a22aabc9bc1ec3b12d68ff2dfcfdb97fd30f4b 100644 (file)
@@ -9,29 +9,6 @@
 #include <linux/dax.h>
 #include <trace/events/erofs.h>
 
-static void erofs_readendio(struct bio *bio)
-{
-       struct bio_vec *bvec;
-       blk_status_t err = bio->bi_status;
-       struct bvec_iter_all iter_all;
-
-       bio_for_each_segment_all(bvec, bio, iter_all) {
-               struct page *page = bvec->bv_page;
-
-               /* page is already locked */
-               DBG_BUGON(PageUptodate(page));
-
-               if (err)
-                       SetPageError(page);
-               else
-                       SetPageUptodate(page);
-
-               unlock_page(page);
-               /* page could be reclaimed now */
-       }
-       bio_put(bio);
-}
-
 struct page *erofs_get_meta_page(struct super_block *sb, erofs_blk_t blkaddr)
 {
        struct address_space *const mapping = sb->s_bdev->bd_inode->i_mapping;
@@ -109,206 +86,6 @@ err_out:
        return err;
 }
 
-static inline struct bio *erofs_read_raw_page(struct bio *bio,
-                                             struct address_space *mapping,
-                                             struct page *page,
-                                             erofs_off_t *last_block,
-                                             unsigned int nblocks,
-                                             unsigned int *eblks,
-                                             bool ra)
-{
-       struct inode *const inode = mapping->host;
-       struct super_block *const sb = inode->i_sb;
-       erofs_off_t current_block = (erofs_off_t)page->index;
-       int err;
-
-       DBG_BUGON(!nblocks);
-
-       if (PageUptodate(page)) {
-               err = 0;
-               goto has_updated;
-       }
-
-       /* note that for readpage case, bio also equals to NULL */
-       if (bio &&
-           (*last_block + 1 != current_block || !*eblks)) {
-submit_bio_retry:
-               submit_bio(bio);
-               bio = NULL;
-       }
-
-       if (!bio) {
-               struct erofs_map_blocks map = {
-                       .m_la = blknr_to_addr(current_block),
-               };
-               erofs_blk_t blknr;
-               unsigned int blkoff;
-
-               err = erofs_map_blocks_flatmode(inode, &map, EROFS_GET_BLOCKS_RAW);
-               if (err)
-                       goto err_out;
-
-               /* zero out the holed page */
-               if (!(map.m_flags & EROFS_MAP_MAPPED)) {
-                       zero_user_segment(page, 0, PAGE_SIZE);
-                       SetPageUptodate(page);
-
-                       /* imply err = 0, see erofs_map_blocks */
-                       goto has_updated;
-               }
-
-               /* for RAW access mode, m_plen must be equal to m_llen */
-               DBG_BUGON(map.m_plen != map.m_llen);
-
-               blknr = erofs_blknr(map.m_pa);
-               blkoff = erofs_blkoff(map.m_pa);
-
-               /* deal with inline page */
-               if (map.m_flags & EROFS_MAP_META) {
-                       void *vsrc, *vto;
-                       struct page *ipage;
-
-                       DBG_BUGON(map.m_plen > PAGE_SIZE);
-
-                       ipage = erofs_get_meta_page(inode->i_sb, blknr);
-
-                       if (IS_ERR(ipage)) {
-                               err = PTR_ERR(ipage);
-                               goto err_out;
-                       }
-
-                       vsrc = kmap_atomic(ipage);
-                       vto = kmap_atomic(page);
-                       memcpy(vto, vsrc + blkoff, map.m_plen);
-                       memset(vto + map.m_plen, 0, PAGE_SIZE - map.m_plen);
-                       kunmap_atomic(vto);
-                       kunmap_atomic(vsrc);
-                       flush_dcache_page(page);
-
-                       SetPageUptodate(page);
-                       /* TODO: could we unlock the page earlier? */
-                       unlock_page(ipage);
-                       put_page(ipage);
-
-                       /* imply err = 0, see erofs_map_blocks */
-                       goto has_updated;
-               }
-
-               /* pa must be block-aligned for raw reading */
-               DBG_BUGON(erofs_blkoff(map.m_pa));
-
-               /* max # of continuous pages */
-               if (nblocks > DIV_ROUND_UP(map.m_plen, PAGE_SIZE))
-                       nblocks = DIV_ROUND_UP(map.m_plen, PAGE_SIZE);
-
-               *eblks = bio_max_segs(nblocks);
-               bio = bio_alloc(GFP_NOIO, *eblks);
-
-               bio->bi_end_io = erofs_readendio;
-               bio_set_dev(bio, sb->s_bdev);
-               bio->bi_iter.bi_sector = (sector_t)blknr <<
-                       LOG_SECTORS_PER_BLOCK;
-               bio->bi_opf = REQ_OP_READ | (ra ? REQ_RAHEAD : 0);
-       }
-
-       err = bio_add_page(bio, page, PAGE_SIZE, 0);
-       /* out of the extent or bio is full */
-       if (err < PAGE_SIZE)
-               goto submit_bio_retry;
-       --*eblks;
-       *last_block = current_block;
-       return bio;
-
-err_out:
-       /* for sync reading, set page error immediately */
-       if (!ra) {
-               SetPageError(page);
-               ClearPageUptodate(page);
-       }
-has_updated:
-       unlock_page(page);
-
-       /* if updated manually, continuous pages has a gap */
-       if (bio)
-               submit_bio(bio);
-       return err ? ERR_PTR(err) : NULL;
-}
-
-/*
- * since we dont have write or truncate flows, so no inode
- * locking needs to be held at the moment.
- */
-static int erofs_raw_access_readpage(struct file *file, struct page *page)
-{
-       erofs_off_t last_block;
-       unsigned int eblks;
-       struct bio *bio;
-
-       trace_erofs_readpage(page, true);
-
-       bio = erofs_read_raw_page(NULL, page->mapping,
-                                 page, &last_block, 1, &eblks, false);
-
-       if (IS_ERR(bio))
-               return PTR_ERR(bio);
-
-       if (bio)
-               submit_bio(bio);
-       return 0;
-}
-
-static void erofs_raw_access_readahead(struct readahead_control *rac)
-{
-       erofs_off_t last_block;
-       unsigned int eblks;
-       struct bio *bio = NULL;
-       struct page *page;
-
-       trace_erofs_readpages(rac->mapping->host, readahead_index(rac),
-                       readahead_count(rac), true);
-
-       while ((page = readahead_page(rac))) {
-               prefetchw(&page->flags);
-
-               bio = erofs_read_raw_page(bio, rac->mapping, page, &last_block,
-                               readahead_count(rac), &eblks, true);
-
-               /* all the page errors are ignored when readahead */
-               if (IS_ERR(bio)) {
-                       pr_err("%s, readahead error at page %lu of nid %llu\n",
-                              __func__, page->index,
-                              EROFS_I(rac->mapping->host)->nid);
-
-                       bio = NULL;
-               }
-
-               put_page(page);
-       }
-
-       if (bio)
-               submit_bio(bio);
-}
-
-static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
-{
-       struct inode *inode = mapping->host;
-       struct erofs_map_blocks map = {
-               .m_la = blknr_to_addr(block),
-       };
-
-       if (EROFS_I(inode)->datalayout == EROFS_INODE_FLAT_INLINE) {
-               erofs_blk_t blks = i_size_read(inode) >> LOG_BLOCK_SIZE;
-
-               if (block >> LOG_SECTORS_PER_BLOCK >= blks)
-                       return 0;
-       }
-
-       if (!erofs_map_blocks_flatmode(inode, &map, EROFS_GET_BLOCKS_RAW))
-               return erofs_blknr(map.m_pa);
-
-       return 0;
-}
-
 static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
                unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
 {
@@ -327,6 +104,7 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
        iomap->offset = map.m_la;
        iomap->length = map.m_llen;
        iomap->flags = 0;
+       iomap->private = NULL;
 
        if (!(map.m_flags & EROFS_MAP_MAPPED)) {
                iomap->type = IOMAP_HOLE;
@@ -336,20 +114,63 @@ static int erofs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
                return 0;
        }
 
-       /* that shouldn't happen for now */
        if (map.m_flags & EROFS_MAP_META) {
-               DBG_BUGON(1);
-               return -ENOTBLK;
+               struct page *ipage;
+
+               iomap->type = IOMAP_INLINE;
+               ipage = erofs_get_meta_page(inode->i_sb,
+                                           erofs_blknr(map.m_pa));
+               if (IS_ERR(ipage))
+                       return PTR_ERR(ipage);
+               iomap->inline_data = page_address(ipage) +
+                                       erofs_blkoff(map.m_pa);
+               iomap->private = ipage;
+       } else {
+               iomap->type = IOMAP_MAPPED;
+               iomap->addr = map.m_pa;
        }
-       iomap->type = IOMAP_MAPPED;
-       iomap->addr = map.m_pa;
        return 0;
 }
 
+static int erofs_iomap_end(struct inode *inode, loff_t pos, loff_t length,
+               ssize_t written, unsigned int flags, struct iomap *iomap)
+{
+       struct page *ipage = iomap->private;
+
+       if (ipage) {
+               DBG_BUGON(iomap->type != IOMAP_INLINE);
+               unlock_page(ipage);
+               put_page(ipage);
+       } else {
+               DBG_BUGON(iomap->type == IOMAP_INLINE);
+       }
+       return written;
+}
+
 static const struct iomap_ops erofs_iomap_ops = {
        .iomap_begin = erofs_iomap_begin,
+       .iomap_end = erofs_iomap_end,
 };
 
+/*
+ * since we dont have write or truncate flows, so no inode
+ * locking needs to be held at the moment.
+ */
+static int erofs_readpage(struct file *file, struct page *page)
+{
+       return iomap_readpage(page, &erofs_iomap_ops);
+}
+
+static void erofs_readahead(struct readahead_control *rac)
+{
+       return iomap_readahead(rac, &erofs_iomap_ops);
+}
+
+static sector_t erofs_bmap(struct address_space *mapping, sector_t block)
+{
+       return iomap_bmap(mapping, block, &erofs_iomap_ops);
+}
+
 static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to)
 {
        struct inode *inode = file_inode(iocb->ki_filp);
@@ -365,15 +186,6 @@ static int erofs_prepare_dio(struct kiocb *iocb, struct iov_iter *to)
 
        if (align & blksize_mask)
                return -EINVAL;
-
-       /*
-        * Temporarily fall back tail-packing inline to buffered I/O instead
-        * since tail-packing inline support relies on an iomap core update.
-        */
-       if (EROFS_I(inode)->datalayout == EROFS_INODE_FLAT_INLINE &&
-           iocb->ki_pos + iov_iter_count(to) >
-                       rounddown(inode->i_size, EROFS_BLKSIZ))
-               return 1;
        return 0;
 }
 
@@ -401,8 +213,8 @@ static ssize_t erofs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 
 /* for uncompressed (aligned) files and raw access for other files */
 const struct address_space_operations erofs_raw_access_aops = {
-       .readpage = erofs_raw_access_readpage,
-       .readahead = erofs_raw_access_readahead,
+       .readpage = erofs_readpage,
+       .readahead = erofs_readahead,
        .bmap = erofs_bmap,
        .direct_IO = noop_direct_IO,
 };