[ 100/101] pnfsblock: fix non-aligned DIO read

[Greg Kroah-Hartman]

3.6-stable review patch.  If anyone has any objections, please let me know.

------------------

From: Peng Tao <bergwolf@xxxxxxxxx>

commit f742dc4a32587bff50b13dde9d8894b96851951a upstream.

For DIO read, if it is not sector aligned, we should reject it
and resend via MDS. Otherwise there might be data corruption.
Also teach bl_read_pagelist to handle partial page reads for DIO.

Signed-off-by: Peng Tao <tao.peng@xxxxxxx>
Signed-off-by: Trond Myklebust <Trond.Myklebust@xxxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>

---
 fs/nfs/blocklayout/blocklayout.c |   77 +++++++++++++++++++++++----------------
 1 file changed, 46 insertions(+), 31 deletions(-)

--- a/fs/nfs/blocklayout/blocklayout.c
+++ b/fs/nfs/blocklayout/blocklayout.c
@@ -242,14 +242,6 @@ bl_end_par_io_read(void *data, int unuse
 	schedule_work(&rdata->task.u.tk_work);
 }
 
-static bool
-bl_check_alignment(u64 offset, u32 len, unsigned long blkmask)
-{
-	if ((offset & blkmask) || (len & blkmask))
-		return false;
-	return true;
-}
-
 static enum pnfs_try_status
 bl_read_pagelist(struct nfs_read_data *rdata)
 {
@@ -260,15 +252,15 @@ bl_read_pagelist(struct nfs_read_data *r
 	sector_t isect, extent_length = 0;
 	struct parallel_io *par;
 	loff_t f_offset = rdata->args.offset;
+	size_t bytes_left = rdata->args.count;
+	unsigned int pg_offset, pg_len;
 	struct page **pages = rdata->args.pages;
 	int pg_index = rdata->args.pgbase >> PAGE_CACHE_SHIFT;
+	const bool is_dio = (header->dreq != NULL);
 
 	dprintk("%s enter nr_pages %u offset %lld count %u\n", __func__,
 	       rdata->pages.npages, f_offset, (unsigned int)rdata->args.count);
 
-	if (!bl_check_alignment(f_offset, rdata->args.count, PAGE_CACHE_MASK))
-		goto use_mds;
-
 	par = alloc_parallel(rdata);
 	if (!par)
 		goto use_mds;
@@ -298,36 +290,53 @@ bl_read_pagelist(struct nfs_read_data *r
 				extent_length = min(extent_length, cow_length);
 			}
 		}
+
+		if (is_dio) {
+			pg_offset = f_offset & ~PAGE_CACHE_MASK;
+			if (pg_offset + bytes_left > PAGE_CACHE_SIZE)
+				pg_len = PAGE_CACHE_SIZE - pg_offset;
+			else
+				pg_len = bytes_left;
+
+			f_offset += pg_len;
+			bytes_left -= pg_len;
+			isect += (pg_offset >> SECTOR_SHIFT);
+		} else {
+			pg_offset = 0;
+			pg_len = PAGE_CACHE_SIZE;
+		}
+
 		hole = is_hole(be, isect);
 		if (hole && !cow_read) {
 			bio = bl_submit_bio(READ, bio);
 			/* Fill hole w/ zeroes w/o accessing device */
 			dprintk("%s Zeroing page for hole\n", __func__);
-			zero_user_segment(pages[i], 0, PAGE_CACHE_SIZE);
+			zero_user_segment(pages[i], pg_offset, pg_len);
 			print_page(pages[i]);
 			SetPageUptodate(pages[i]);
 		} else {
 			struct pnfs_block_extent *be_read;
 
 			be_read = (hole && cow_read) ? cow_read : be;
-			bio = bl_add_page_to_bio(bio, rdata->pages.npages - i,
+			bio = do_add_page_to_bio(bio, rdata->pages.npages - i,
 						 READ,
 						 isect, pages[i], be_read,
-						 bl_end_io_read, par);
+						 bl_end_io_read, par,
+						 pg_offset, pg_len);
 			if (IS_ERR(bio)) {
 				header->pnfs_error = PTR_ERR(bio);
 				bio = NULL;
 				goto out;
 			}
 		}
-		isect += PAGE_CACHE_SECTORS;
+		isect += (pg_len >> SECTOR_SHIFT);
 		extent_length -= PAGE_CACHE_SECTORS;
 	}
 	if ((isect << SECTOR_SHIFT) >= header->inode->i_size) {
 		rdata->res.eof = 1;
-		rdata->res.count = header->inode->i_size - f_offset;
+		rdata->res.count = header->inode->i_size - rdata->args.offset;
 	} else {
-		rdata->res.count = (isect << SECTOR_SHIFT) - f_offset;
+		rdata->res.count = (isect << SECTOR_SHIFT) - rdata->args.offset;
 	}
 out:
 	bl_put_extent(be);
@@ -676,7 +685,7 @@ bl_write_pagelist(struct nfs_write_data
 	struct bio *bio = NULL;
 	struct pnfs_block_extent *be = NULL, *cow_read = NULL;
 	sector_t isect, last_isect = 0, extent_length = 0;
-	struct parallel_io *par = NULL;
+	struct parallel_io *par;
 	loff_t offset = wdata->args.offset;
 	size_t count = wdata->args.count;
 	unsigned int pg_offset, pg_len, saved_len;
@@ -688,10 +697,6 @@ bl_write_pagelist(struct nfs_write_data
 	    NFS_SERVER(header->inode)->pnfs_blksize >> PAGE_CACHE_SHIFT;
 
 	dprintk("%s enter, %Zu@%lld\n", __func__, count, offset);
-	/* Check for alignment first */
-	if (!bl_check_alignment(offset, count, PAGE_CACHE_MASK))
-		goto out_mds;
-
 	/* At this point, wdata->pages is a (sequential) list of nfs_pages.
 	 * We want to write each, and if there is an error set pnfs_error
 	 * to have it redone using nfs.
@@ -1164,32 +1169,42 @@ bl_clear_layoutdriver(struct nfs_server
 	return 0;
 }
 
+static bool
+is_aligned_req(struct nfs_page *req, unsigned int alignment)
+{
+	return IS_ALIGNED(req->wb_offset, alignment) &&
+	       IS_ALIGNED(req->wb_bytes, alignment);
+}
+
 static void
 bl_pg_init_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
 {
-	if (!bl_check_alignment(req->wb_offset, req->wb_bytes, PAGE_CACHE_MASK))
+	if (pgio->pg_dreq != NULL &&
+	    !is_aligned_req(req, SECTOR_SIZE))
 		nfs_pageio_reset_read_mds(pgio);
 	else
 		pnfs_generic_pg_init_read(pgio, req);
 }
 
-static void
-bl_pg_init_write(struct nfs_pageio_descriptor *pgio, struct nfs_page *req)
+static bool
+bl_pg_test_read(struct nfs_pageio_descriptor *pgio, struct nfs_page *prev,
+		struct nfs_page *req)
 {
-	if (!bl_check_alignment(req->wb_offset, req->wb_bytes, PAGE_CACHE_MASK))
-		nfs_pageio_reset_write_mds(pgio);
-	else
-		pnfs_generic_pg_init_write(pgio, req);
+	if (pgio->pg_dreq != NULL &&
+	    !is_aligned_req(req, SECTOR_SIZE))
+		return false;
+
+	return pnfs_generic_pg_test(pgio, prev, req);
 }
 
 static const struct nfs_pageio_ops bl_pg_read_ops = {
 	.pg_init = bl_pg_init_read,
-	.pg_test = pnfs_generic_pg_test,
+	.pg_test = bl_pg_test_read,
 	.pg_doio = pnfs_generic_pg_readpages,
 };
 
 static const struct nfs_pageio_ops bl_pg_write_ops = {
-	.pg_init = bl_pg_init_write,
+	.pg_init = pnfs_generic_pg_init_write,
 	.pg_test = pnfs_generic_pg_test,
 	.pg_doio = pnfs_generic_pg_writepages,
 };


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/