[Iscsitarget-commits] SF.net SVN: iscsitarget:[494] trunk/kernel

[<ag...@us...>]

Revision: 494
          http://sourceforge.net/p/iscsitarget/code/494
Author:   agr1
Date:     2013-07-03 18:34:49 +0000 (Wed, 03 Jul 2013)
Log Message:
-----------
Avoid zero sized bio_add_page (blockio) / vfs_{read,write} (fileio) calls

From: Shivaram Upadhyayula <shi...@qu...>

Let suppose the total write length for the write same command is 1048576 + 512 (1MB + 512).
As per the logic the writes are split into two due to the MAX_IO_SIZE limitation of 1MB
tio pg_cnt will be (1MB / 4096)) == 256

        while (length > 0) {
                u32 to_write = (u32)min_t(u64, length, MAX_IO_SIZE);

So the first iteration will be to_write == 1M
The second iteration will be to_write == 512
target_tio pg_cnt will be 256 as we dont modify that and is required to free up the pages later

Now examining block-io.c make_request

        while (tio_index < tio->pg_cnt) {

Notice above that for the second iteration we will iterate from 0 to 256 but we need to write only 512 bytes.
With the 512 byte write, for the first iteration in the below loop

                /* Loop for filling bio */
                while (tio_index < tio->pg_cnt) {

                        unsigned int bytes = PAGE_SIZE;

                        if (bytes > size)
                                bytes = size;

                        if (!bio_add_page(bio, tio->pvec[tio_index], bytes, 0))
                                break;

                        size -= bytes;
                        ppos += bytes;

                        tio_index++;

In the above loop, we add the 512 bytes to the bio in the first iteration itself. In the second iteration of the
above loops bytes == size == 0 and so on till 256 and in the second iteration we call bio_add_page with size 0.

(redlicha: this was part of another patch and split out to a distinct patch)

Signed-off-by: Shivaram Upadhyayula <shi...@qu...>
Signed-off-by: Arne Redlich <arn...@go...>

Modified Paths:
--------------
    trunk/kernel/block-io.c
    trunk/kernel/file-io.c

Modified: trunk/kernel/block-io.c
===================================================================
--- trunk/kernel/block-io.c	2013-07-02 19:33:19 UTC (rev 493)
+++ trunk/kernel/block-io.c	2013-07-03 18:34:49 UTC (rev 494)
@@ -79,7 +79,7 @@
 	init_completion(&tio_work->tio_complete);
 
 	/* Main processing loop, allocate and fill all bios */
-	while (tio_index < tio->pg_cnt) {
+	while (size && tio_index < tio->pg_cnt) {
 		bio = bio_alloc(GFP_KERNEL, min(max_pages, BIO_MAX_PAGES));
 		if (!bio) {
 			err = -ENOMEM;
@@ -100,7 +100,7 @@
 		atomic_inc(&tio_work->bios_remaining);
 
 		/* Loop for filling bio */
-		while (tio_index < tio->pg_cnt) {
+		while (size && tio_index < tio->pg_cnt) {
 			unsigned int bytes = PAGE_SIZE;
 
 			if (bytes > size)

Modified: trunk/kernel/file-io.c
===================================================================
--- trunk/kernel/file-io.c	2013-07-02 19:33:19 UTC (rev 493)
+++ trunk/kernel/file-io.c	2013-07-03 18:34:49 UTC (rev 494)
@@ -34,7 +34,7 @@
 	size = tio->size;
 	ppos = tio->offset;
 
-	for (i = 0; i < tio->pg_cnt; i++) {
+	for (i = 0; i < tio->pg_cnt && size; i++) {
 		page = tio->pvec[i];
 		assert(page);
 		buf = page_address(page);

This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.