OSDN Git Service

iomap: add an iomap-based readpage and readpages implementation
authorChristoph Hellwig <hch@lst.de>
Tue, 19 Jun 2018 22:10:57 +0000 (15:10 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Tue, 19 Jun 2018 22:10:57 +0000 (15:10 -0700)
Simply use iomap_apply to iterate over the file and a submit a bio for
each non-uptodate but mapped region and zero everything else.  Note that
as-is this can not be used for file systems with a blocksize smaller than
the page size, but that support will be added later.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
fs/iomap.c
include/linux/iomap.h

index a1f71e6..4f10c6b 100644 (file)
@@ -1,6 +1,6 @@
 /*
  * Copyright (C) 2010 Red Hat, Inc.
- * Copyright (c) 2016 Christoph Hellwig.
+ * Copyright (c) 2016-2018 Christoph Hellwig.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms and conditions of the GNU General Public License,
@@ -18,6 +18,7 @@
 #include <linux/uaccess.h>
 #include <linux/gfp.h>
 #include <linux/mm.h>
+#include <linux/mm_inline.h>
 #include <linux/swap.h>
 #include <linux/pagemap.h>
 #include <linux/pagevec.h>
@@ -124,6 +125,217 @@ iomap_read_inline_data(struct inode *inode, struct page *page,
 }
 
 static void
+iomap_read_end_io(struct bio *bio)
+{
+       int error = blk_status_to_errno(bio->bi_status);
+       struct bio_vec *bvec;
+       int i;
+
+       bio_for_each_segment_all(bvec, bio, i)
+               page_endio(bvec->bv_page, false, error);
+       bio_put(bio);
+}
+
+struct iomap_readpage_ctx {
+       struct page             *cur_page;
+       bool                    cur_page_in_bio;
+       bool                    is_readahead;
+       struct bio              *bio;
+       struct list_head        *pages;
+};
+
+static loff_t
+iomap_readpage_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
+               struct iomap *iomap)
+{
+       struct iomap_readpage_ctx *ctx = data;
+       struct page *page = ctx->cur_page;
+       unsigned poff = pos & (PAGE_SIZE - 1);
+       unsigned plen = min_t(loff_t, PAGE_SIZE - poff, length);
+       bool is_contig = false;
+       sector_t sector;
+
+       /* we don't support blocksize < PAGE_SIZE quite yet. */
+       WARN_ON_ONCE(pos != page_offset(page));
+       WARN_ON_ONCE(plen != PAGE_SIZE);
+
+       if (iomap->type != IOMAP_MAPPED || pos >= i_size_read(inode)) {
+               zero_user(page, poff, plen);
+               SetPageUptodate(page);
+               goto done;
+       }
+
+       ctx->cur_page_in_bio = true;
+
+       /*
+        * Try to merge into a previous segment if we can.
+        */
+       sector = iomap_sector(iomap, pos);
+       if (ctx->bio && bio_end_sector(ctx->bio) == sector) {
+               if (__bio_try_merge_page(ctx->bio, page, plen, poff))
+                       goto done;
+               is_contig = true;
+       }
+
+       if (!ctx->bio || !is_contig || bio_full(ctx->bio)) {
+               gfp_t gfp = mapping_gfp_constraint(page->mapping, GFP_KERNEL);
+               int nr_vecs = (length + PAGE_SIZE - 1) >> PAGE_SHIFT;
+
+               if (ctx->bio)
+                       submit_bio(ctx->bio);
+
+               if (ctx->is_readahead) /* same as readahead_gfp_mask */
+                       gfp |= __GFP_NORETRY | __GFP_NOWARN;
+               ctx->bio = bio_alloc(gfp, min(BIO_MAX_PAGES, nr_vecs));
+               ctx->bio->bi_opf = REQ_OP_READ;
+               if (ctx->is_readahead)
+                       ctx->bio->bi_opf |= REQ_RAHEAD;
+               ctx->bio->bi_iter.bi_sector = sector;
+               bio_set_dev(ctx->bio, iomap->bdev);
+               ctx->bio->bi_end_io = iomap_read_end_io;
+       }
+
+       __bio_add_page(ctx->bio, page, plen, poff);
+done:
+       return plen;
+}
+
+int
+iomap_readpage(struct page *page, const struct iomap_ops *ops)
+{
+       struct iomap_readpage_ctx ctx = { .cur_page = page };
+       struct inode *inode = page->mapping->host;
+       unsigned poff;
+       loff_t ret;
+
+       WARN_ON_ONCE(page_has_buffers(page));
+
+       for (poff = 0; poff < PAGE_SIZE; poff += ret) {
+               ret = iomap_apply(inode, page_offset(page) + poff,
+                               PAGE_SIZE - poff, 0, ops, &ctx,
+                               iomap_readpage_actor);
+               if (ret <= 0) {
+                       WARN_ON_ONCE(ret == 0);
+                       SetPageError(page);
+                       break;
+               }
+       }
+
+       if (ctx.bio) {
+               submit_bio(ctx.bio);
+               WARN_ON_ONCE(!ctx.cur_page_in_bio);
+       } else {
+               WARN_ON_ONCE(ctx.cur_page_in_bio);
+               unlock_page(page);
+       }
+
+       /*
+        * Just like mpage_readpages and block_read_full_page we always
+        * return 0 and just mark the page as PageError on errors.  This
+        * should be cleaned up all through the stack eventually.
+        */
+       return 0;
+}
+EXPORT_SYMBOL_GPL(iomap_readpage);
+
+static struct page *
+iomap_next_page(struct inode *inode, struct list_head *pages, loff_t pos,
+               loff_t length, loff_t *done)
+{
+       while (!list_empty(pages)) {
+               struct page *page = lru_to_page(pages);
+
+               if (page_offset(page) >= (u64)pos + length)
+                       break;
+
+               list_del(&page->lru);
+               if (!add_to_page_cache_lru(page, inode->i_mapping, page->index,
+                               GFP_NOFS))
+                       return page;
+
+               /*
+                * If we already have a page in the page cache at index we are
+                * done.  Upper layers don't care if it is uptodate after the
+                * readpages call itself as every page gets checked again once
+                * actually needed.
+                */
+               *done += PAGE_SIZE;
+               put_page(page);
+       }
+
+       return NULL;
+}
+
+static loff_t
+iomap_readpages_actor(struct inode *inode, loff_t pos, loff_t length,
+               void *data, struct iomap *iomap)
+{
+       struct iomap_readpage_ctx *ctx = data;
+       loff_t done, ret;
+
+       for (done = 0; done < length; done += ret) {
+               if (ctx->cur_page && ((pos + done) & (PAGE_SIZE - 1)) == 0) {
+                       if (!ctx->cur_page_in_bio)
+                               unlock_page(ctx->cur_page);
+                       put_page(ctx->cur_page);
+                       ctx->cur_page = NULL;
+               }
+               if (!ctx->cur_page) {
+                       ctx->cur_page = iomap_next_page(inode, ctx->pages,
+                                       pos, length, &done);
+                       if (!ctx->cur_page)
+                               break;
+                       ctx->cur_page_in_bio = false;
+               }
+               ret = iomap_readpage_actor(inode, pos + done, length - done,
+                               ctx, iomap);
+       }
+
+       return done;
+}
+
+int
+iomap_readpages(struct address_space *mapping, struct list_head *pages,
+               unsigned nr_pages, const struct iomap_ops *ops)
+{
+       struct iomap_readpage_ctx ctx = {
+               .pages          = pages,
+               .is_readahead   = true,
+       };
+       loff_t pos = page_offset(list_entry(pages->prev, struct page, lru));
+       loff_t last = page_offset(list_entry(pages->next, struct page, lru));
+       loff_t length = last - pos + PAGE_SIZE, ret = 0;
+
+       while (length > 0) {
+               ret = iomap_apply(mapping->host, pos, length, 0, ops,
+                               &ctx, iomap_readpages_actor);
+               if (ret <= 0) {
+                       WARN_ON_ONCE(ret == 0);
+                       goto done;
+               }
+               pos += ret;
+               length -= ret;
+       }
+       ret = 0;
+done:
+       if (ctx.bio)
+               submit_bio(ctx.bio);
+       if (ctx.cur_page) {
+               if (!ctx.cur_page_in_bio)
+                       unlock_page(ctx.cur_page);
+               put_page(ctx.cur_page);
+       }
+
+       /*
+        * Check that we didn't lose a page due to the arcance calling
+        * conventions..
+        */
+       WARN_ON_ONCE(!ret && !list_empty(ctx.pages));
+       return ret;
+}
+EXPORT_SYMBOL_GPL(iomap_readpages);
+
+static void
 iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
 {
        loff_t i_size = i_size_read(inode);
index 1f36523..30d3144 100644 (file)
@@ -99,6 +99,9 @@ struct iomap_ops {
 
 ssize_t iomap_file_buffered_write(struct kiocb *iocb, struct iov_iter *from,
                const struct iomap_ops *ops);
+int iomap_readpage(struct page *page, const struct iomap_ops *ops);
+int iomap_readpages(struct address_space *mapping, struct list_head *pages,
+               unsigned nr_pages, const struct iomap_ops *ops);
 int iomap_file_dirty(struct inode *inode, loff_t pos, loff_t len,
                const struct iomap_ops *ops);
 int iomap_zero_range(struct inode *inode, loff_t pos, loff_t len,