md/raid1:If r1bio->sectors % 8 != 0,then the memcmp and later memcpy will omit the last bio_vec.

PATCH majianpeng 3 participants 4 messages Mar 31, 2012 – Apr 5, 2012 MD RAID

Share & tools

majianpeng

author

2012-03-31 07:44 UTC #0 lore raw

From 72e5ad9da511689f7efe330ce5093a3df2c92738 Mon Sep 17 00:00:00 2001
From: majianpeng <[email protected]>
Date: Sat, 31 Mar 2012 15:36:01 +0800
Subject: [PATCH] md/raid1:If r1bio->sectors % 8 != 0,then the memcmp and
later memcpy will omit the last bio_vec.

Signed-off-by majianpeng <[email protected]>

---
 drivers/md/raid1.c |   14 ++++++++++----
 1 files changed, 10 insertions(+), 4 deletions(-)

drivers/md/raid1.c +10 -4

index 4a40a20..11a28ca 100644
@@ -1711,7 +1711,7 @@ static int process_checks(struct r1bio *r1_bio)
 	struct mddev *mddev = r1_bio->mddev;
 	struct r1conf *conf = mddev->private;
 	int primary;
-	int i;
+	int i, vcnt;
 
 	for (primary = 0; primary < conf->raid_disks * 2; primary++)
 		if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&

@@ -1721,9 +1721,9 @@ static int process_checks(struct r1bio *r1_bio)
 			break;
 		}
 	r1_bio->read_disk = primary;
+	vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
 	for (i = 0; i < conf->raid_disks * 2; i++) {
 		int j;
-		int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
 		struct bio *pbio = r1_bio->bios[primary];
 		struct bio *sbio = r1_bio->bios[i];
 		int size;

@@ -1736,9 +1736,16 @@ static int process_checks(struct r1bio *r1_bio)
 				struct page *p, *s;
 				p = pbio->bi_io_vec[j].bv_page;
 				s = sbio->bi_io_vec[j].bv_page;
+				if ((j == vcnt - 1) &&
+					(r1_bio->sectors &
+					 (PAGE_SIZE / 512 - 1)))
+					size = (r1_bio->sectors << 9)
+						- (PAGE_SIZE * j);
+				else
+					size = PAGE_SIZE;
 				if (memcmp(page_address(p),
 					   page_address(s),
-					   PAGE_SIZE))
+					   size))
 					break;
 			}
 		} else

@@ -2480,7 +2487,6 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipp
 	} while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
  bio_full:
 	r1_bio->sectors = nr_sectors;
-
 	/* For a user-requested sync, we read all readable devices and do a
 	 * compare
 	 */

Signature · 6 lines

1.7.5.4

 				
--------------
majianpeng
2012-03-31

NeilBrown

2012-04-02 01:41 UTC #1 lore raw

Quoted 5 lines · On Sat, 31 Mar 2012 15:44:14 +0800 "majianpeng" <[email protected]> wrote:

Quoted 1 line
From 72e5ad9da511689f7efe330ce5093a3df2c92738 Mon Sep 17 00:00:00 2001
From: majianpeng <[email protected]>
Date: Sat, 31 Mar 2012 15:36:01 +0800
Subject: [PATCH] md/raid1:If r1bio->sectors % 8 != 0,then the memcmp and
later memcpy will omit the last bio_vec.

Thanks for this.

There is a bug here. However I think your fix is overly complex.

Also the 'memcpy' doesn't need to be changed - it doesn't hurt if it copies more bytes than necessary.
However it could hurt if we compare more bytes than necessary.

I've applied a very different patch as below.

Thanks,
NeilBrown

From 9d49d24d665146d121d3ae349adc92e83460d75d Mon Sep 17 00:00:00 2001
From: NeilBrown <[email protected]>
Date: Mon, 2 Apr 2012 11:39:05 +1000
Subject: [PATCH] md/raid1,raid10: don't compare excess byte during
consistency check.

When comparing two pages read from different legs of a mirror, only compare the bytes that were read, not the whole page.

I most cases were read a whole page, but in some cases with bad blocks or odd sizes devices we might read fewer than that.

This bug has been present "forever" but at worst it might cause a report of two many mismatches and generate a little bit extra resync IO, so there is no need to back-port to -stable kernels.

Reported-by majianpeng <[email protected]>

Signed-off-by NeilBrown <[email protected]>

drivers/md/raid1.c +1 -1

index 8c420f1..d35e4c9 100644
@@ -1738,7 +1738,7 @@ static int process_checks(struct r1bio *r1_bio)
 				s = sbio->bi_io_vec[j].bv_page;
 				if (memcmp(page_address(p),
 					   page_address(s),
-					   PAGE_SIZE))
+					   sbio->bi_io_vec[j].bv_len))
 					break;
 			}
 		} else

drivers/md/raid10.c +1 -1

index 3540316..fff7821 100644
@@ -1821,7 +1821,7 @@ static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio)
 			for (j = 0; j < vcnt; j++)
 				if (memcmp(page_address(fbio->bi_io_vec[j].bv_page),
 					   page_address(tbio->bi_io_vec[j].bv_page),
-					   PAGE_SIZE))
+					   fbio->bi_io_vec[j].bv_len))
 					break;
 			if (j == vcnt)
 				continue;

Quoted 57 lines

Signed-off-by majianpeng <[email protected]>

---
 drivers/md/raid1.c |   14 ++++++++++----
 1 files changed, 10 insertions(+), 4 deletions(-)

drivers/md/raid1.c +10 -4

index 4a40a20..11a28ca 100644
@@ -1711,7 +1711,7 @@ static int process_checks(struct r1bio *r1_bio)
 	struct mddev *mddev = r1_bio->mddev;
 	struct r1conf *conf = mddev->private;
 	int primary;
-	int i;
+	int i, vcnt;
 
 	for (primary = 0; primary < conf->raid_disks * 2; primary++)
 		if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&

@@ -1721,9 +1721,9 @@ static int process_checks(struct r1bio *r1_bio)
 			break;
 		}
 	r1_bio->read_disk = primary;
+	vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
 	for (i = 0; i < conf->raid_disks * 2; i++) {
 		int j;
-		int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
 		struct bio *pbio = r1_bio->bios[primary];
 		struct bio *sbio = r1_bio->bios[i];
 		int size;

@@ -1736,9 +1736,16 @@ static int process_checks(struct r1bio *r1_bio)
 				struct page *p, *s;
 				p = pbio->bi_io_vec[j].bv_page;
 				s = sbio->bi_io_vec[j].bv_page;
+				if ((j == vcnt - 1) &&
+					(r1_bio->sectors &
+					 (PAGE_SIZE / 512 - 1)))
+					size = (r1_bio->sectors << 9)
+						- (PAGE_SIZE * j);
+				else
+					size = PAGE_SIZE;
 				if (memcmp(page_address(p),
 					   page_address(s),
-					   PAGE_SIZE))
+					   size))
 					break;
 			}
 		} else

@@ -2480,7 +2487,6 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipp
 	} while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
  bio_full:
 	r1_bio->sectors = nr_sectors;
-
 	/* For a user-requested sync, we read all readable devices and do a
 	 * compare
 	 */

kedacomkernel

2012-04-05 06:12 UTC #2 lore raw

Re: Re: md/raid1:If r1bio->sectors % 8 != 0,then the memcmp and latermemcpy will omit the last bio_vec.

In reply to majianpeng

I readed your patch and found it wrong.My patch had two point 1:memcmp probem.You patch can correct it.
2:omit the last bio_vec.You patch did not correct it.
If r1bio->sectors %8 != 0,supposed sectors = 18,then bio cover three pages. But:

	>>int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9); vcnt=2

then

Quoted 1 line

Quoted 1 line
for (j = vcnt; j-- ; )

      j = 1, 0

------------------
kedacomkernel
2012-04-05

-------------------------------------------------------------
发件人：NeilBrown
发送日期：2012-04-02 09:41:29
收件人：majianpeng
抄送：linux-raid
主题：Re: md/raid1:If r1bio->sectors % 8 != 0,then the memcmp and latermemcpy will omit the last bio_vec.

Quoted 5 lines · On Sat, 31 Mar 2012 15:44:14 +0800 "majianpeng" <[email protected]> wrote:

Quoted 1 line
From 72e5ad9da511689f7efe330ce5093a3df2c92738 Mon Sep 17 00:00:00 2001
From: majianpeng <[email protected]>
Date: Sat, 31 Mar 2012 15:36:01 +0800
Subject: [PATCH] md/raid1:If r1bio->sectors % 8 != 0,then the memcmp and
later memcpy will omit the last bio_vec.

Thanks for this.

There is a bug here. However I think your fix is overly complex.

Also the 'memcpy' doesn't need to be changed - it doesn't hurt if it copies more bytes than necessary.
However it could hurt if we compare more bytes than necessary.

I've applied a very different patch as below.

Thanks,
NeilBrown

When comparing two pages read from different legs of a mirror, only compare the bytes that were read, not the whole page.

I most cases were read a whole page, but in some cases with bad blocks or odd sizes devices we might read fewer than that.

This bug has been present "forever" but at worst it might cause a report of two many mismatches and generate a little bit extra resync IO, so there is no need to back-port to -stable kernels.

Reported-by majianpeng <[email protected]>

Signed-off-by NeilBrown <[email protected]>

drivers/md/raid1.c +1 -1

index 8c420f1..d35e4c9 100644
@@ -1738,7 +1738,7 @@ static int process_checks(struct r1bio *r1_bio)
 				s = sbio->bi_io_vec[j].bv_page;
 				if (memcmp(page_address(p),
 					   page_address(s),
-					   PAGE_SIZE))
+					   sbio->bi_io_vec[j].bv_len))
 					break;
 			}
 		} else

drivers/md/raid10.c +1 -1

index 3540316..fff7821 100644
@@ -1821,7 +1821,7 @@ static void sync_request_write(struct mddev *mddev, struct r10bio *r10_bio)
 			for (j = 0; j < vcnt; j++)
 				if (memcmp(page_address(fbio->bi_io_vec[j].bv_page),
 					   page_address(tbio->bi_io_vec[j].bv_page),
-					   PAGE_SIZE))
+					   fbio->bi_io_vec[j].bv_len))
 					break;
 			if (j == vcnt)
 				continue;

Quoted 57 lines

Signed-off-by majianpeng <[email protected]>

---
 drivers/md/raid1.c |   14 ++++++++++----
 1 files changed, 10 insertions(+), 4 deletions(-)

drivers/md/raid1.c +10 -4

index 4a40a20..11a28ca 100644
@@ -1711,7 +1711,7 @@ static int process_checks(struct r1bio *r1_bio)
 	struct mddev *mddev = r1_bio->mddev;
 	struct r1conf *conf = mddev->private;
 	int primary;
-	int i;
+	int i, vcnt;
 
 	for (primary = 0; primary < conf->raid_disks * 2; primary++)
 		if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&

@@ -1721,9 +1721,9 @@ static int process_checks(struct r1bio *r1_bio)
 			break;
 		}
 	r1_bio->read_disk = primary;
+	vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
 	for (i = 0; i < conf->raid_disks * 2; i++) {
 		int j;
-		int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
 		struct bio *pbio = r1_bio->bios[primary];
 		struct bio *sbio = r1_bio->bios[i];
 		int size;

@@ -1736,9 +1736,16 @@ static int process_checks(struct r1bio *r1_bio)
 				struct page *p, *s;
 				p = pbio->bi_io_vec[j].bv_page;
 				s = sbio->bi_io_vec[j].bv_page;
+				if ((j == vcnt - 1) &&
+					(r1_bio->sectors &
+					 (PAGE_SIZE / 512 - 1)))
+					size = (r1_bio->sectors << 9)
+						- (PAGE_SIZE * j);
+				else
+					size = PAGE_SIZE;
 				if (memcmp(page_address(p),
 					   page_address(s),
-					   PAGE_SIZE))
+					   size))
 					break;
 			}
 		} else

@@ -2480,7 +2487,6 @@ static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipp
 	} while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
  bio_full:
 	r1_bio->sectors = nr_sectors;
-
 	/* For a user-requested sync, we read all readable devices and do a
 	 * compare
 	 */

NeilBrown

2012-04-05 06:53 UTC #3 lore raw

Re: md/raid1:If r1bio->sectors % 8 != 0,then the memcmp and latermemcpy will omit the last bio_vec.

Quoted 10 lines · On Thu, 5 Apr 2012 14:12:25 +0800 "kedacomkernel" <[email protected]>…

I readed your patch and found it wrong.My patch had two point 1:memcmp probem.You patch can correct it.
2:omit the last bio_vec.You patch did not correct it.
If r1bio->sectors %8 != 0,supposed sectors = 18,then bio cover three pages. But:
	>>int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9); vcnt=2
then
Quoted 1 line
Quoted 1 line
for (j = vcnt; j-- ; )
      j = 1, 0

Ahhh, yes of course.
Thanks.

Quoted 7 lines

(quoted diff) +1 -1

@@ -1711,7 +1711,7 @@ static int process_checks(struct r1bio *r1_bio)
 	struct mddev *mddev = r1_bio->mddev;
 	struct r1conf *conf = mddev->private;
 	int primary;
-	int i;
+	int i, vcnt;

Quoted 13 lines

Review static int process_checks(struct r1bio *r1_bio) · line 1729

Quoted 13 lines

(quoted diff) +1 -1

 	for (primary = 0; primary < conf->raid_disks * 2; primary++)
 		if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&

@@ -1721,9 +1721,9 @@ static int process_checks(struct r1bio *r1_bio)
 			break;
 		}
 	r1_bio->read_disk = primary;
+	vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
 	for (i = 0; i < conf->raid_disks * 2; i++) {
 		int j;
-		int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
 		struct bio *pbio = r1_bio->bios[primary];
 		struct bio *sbio = r1_bio->bios[i];
 		int size;

This bit.

I'll make sure that gets applied.

Thanks,
NeilBrown