Skip to contentSkip to conversation

block: read i_size with i_size_read()

PATCH In review Mike Snitzer 4 participants 4 messages Nov 5, 2010 – Nov 7, 2010

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 block: read i_size with i_size_read()

Convert direct reads of an inode's i_size to using i_size_read().

i_size_{read,write} use a seqcount to protect reads from accessing incomple writes. Concurrent i_size_write()s require mutual exclussion to protect the seqcount that is used by i_size_{read,write}. But i_size_read() callers do not need to use additional locking.

---
 block/blk-core.c              |    4 ++--
 block/compat_ioctl.c          |    4 ++--
 block/ioctl.c                 |    6 +++---
 drivers/block/drbd/drbd_int.h |    2 +-
 drivers/md/md.c               |   18 +++++++++---------
 5 files changed, 17 insertions(+), 17 deletions(-)

NOTE: I stopped short of fixing drbd and nbd to use revalidate_disk and/or i_size_write. I also have no intention of auditing all the direct i_size readers/writers in fs/, maybe others will have an interest.

block/blk-core.c +2 -2
index f0834e2..17fcb83 100644
@@ -1351,7 +1351,7 @@ static void handle_bad_sector(struct bio *bio)
 			bdevname(bio->bi_bdev, b),
 			bio->bi_rw,
 			(unsigned long long)bio->bi_sector + bio_sectors(bio),
-			(long long)(bio->bi_bdev->bd_inode->i_size >> 9));
+			(long long)(i_size_read(bio->bi_bdev->bd_inode) >> 9));
 
 	set_bit(BIO_EOF, &bio->bi_flags);
 }
@@ -1404,7 +1404,7 @@ static inline int bio_check_eod(struct bio *bio, unsigned int nr_sectors)
 		return 0;
 
 	/* Test device or partition size, when known. */
-	maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
+	maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
 	if (maxsector) {
 		sector_t sector = bio->bi_sector;
 
block/compat_ioctl.c +2 -2
index 119f07b..58c6ee5 100644
@@ -744,13 +744,13 @@ long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 		bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE;
 		return 0;
 	case BLKGETSIZE:
-		size = bdev->bd_inode->i_size;
+		size = i_size_read(bdev->bd_inode);
 		if ((size >> 9) > ~0UL)
 			return -EFBIG;
 		return compat_put_ulong(arg, size >> 9);
 
 	case BLKGETSIZE64_32:
-		return compat_put_u64(arg, bdev->bd_inode->i_size);
+		return compat_put_u64(arg, i_size_read(bdev->bd_inode));
 
 	case BLKTRACESETUP32:
 	case BLKTRACESTART: /* compatible */
block/ioctl.c +3 -3
index d724ceb..38aa194 100644
@@ -125,7 +125,7 @@ static int blk_ioctl_discard(struct block_device *bdev, uint64_t start,
 	start >>= 9;
 	len >>= 9;
 
-	if (start + len > (bdev->bd_inode->i_size >> 9))
+	if (start + len > (i_size_read(bdev->bd_inode) >> 9))
 		return -EINVAL;
 	if (secure)
 		flags |= BLKDEV_DISCARD_SECURE;
@@ -307,12 +307,12 @@ int blkdev_ioctl(struct block_device *bdev, fmode_t mode, unsigned cmd,
 		ret = blkdev_reread_part(bdev);
 		break;
 	case BLKGETSIZE:
-		size = bdev->bd_inode->i_size;
+		size = i_size_read(bdev->bd_inode);
 		if ((size >> 9) > ~0UL)
 			return -EFBIG;
 		return put_ulong(arg, size >> 9);
 	case BLKGETSIZE64:
-		return put_u64(arg, bdev->bd_inode->i_size);
+		return put_u64(arg, i_size_read(bdev->bd_inode));
 	case BLKTRACESTART:
 	case BLKTRACESTOP:
 	case BLKTRACESETUP:
drivers/block/drbd/drbd_int.h +1 -1
index 9bdcf43..2637f49 100644
@@ -1874,7 +1874,7 @@ static inline sector_t drbd_md_last_sector(struct drbd_backing_dev *bdev)
 static inline sector_t drbd_get_capacity(struct block_device *bdev)
 {
 	/* return bdev ? get_capacity(bdev->bd_disk) : 0; */
-	return bdev ? bdev->bd_inode->i_size >> 9 : 0;
+	return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0;
 }
 
 /**
drivers/md/md.c +9 -9
index 4e957f3..02607b1 100644
@@ -706,7 +706,7 @@ static struct mdk_personality *find_pers(int level, char *clevel)
 /* return the offset of the super block in 512byte sectors */
 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 {
-	sector_t num_sectors = bdev->bd_inode->i_size / 512;
+	sector_t num_sectors = i_size_read(bdev->bd_inode) / 512;
 	return MD_NEW_SIZE_SECTORS(num_sectors);
 }
 
@@ -1386,7 +1386,7 @@ static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 	 */
 	switch(minor_version) {
 	case 0:
-		sb_start = rdev->bdev->bd_inode->i_size >> 9;
+		sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
 		sb_start -= 8*2;
 		sb_start &= ~(sector_t)(4*2-1);
 		break;
@@ -1472,7 +1472,7 @@ static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 			ret = 0;
 	}
 	if (minor_version)
-		rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
+		rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
 			le64_to_cpu(sb->data_offset);
 	else
 		rdev->sectors = rdev->sb_start;
@@ -1680,7 +1680,7 @@ super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
 		return 0; /* component must fit device */
 	if (rdev->sb_start < rdev->data_offset) {
 		/* minor versions 1 and 2; superblock before data */
-		max_sectors = rdev->bdev->bd_inode->i_size >> 9;
+		max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
 		max_sectors -= rdev->data_offset;
 		if (!num_sectors || num_sectors > max_sectors)
 			num_sectors = max_sectors;
@@ -1690,7 +1690,7 @@ super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
 	} else {
 		/* minor version 0; superblock after data */
 		sector_t sb_start;
-		sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
+		sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
 		sb_start &= ~(sector_t)(4*2 - 1);
 		max_sectors = rdev->sectors + sb_start - rdev->sb_start;
 		if (!num_sectors || num_sectors > max_sectors)
@@ -2584,7 +2584,7 @@ rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
 			if (!sectors)
 				return -EBUSY;
 		} else if (!sectors)
-			sectors = (rdev->bdev->bd_inode->i_size >> 9) -
+			sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
 				rdev->data_offset;
 	}
 	if (sectors < my_mddev->dev_sectors)
@@ -2797,7 +2797,7 @@ static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_mi
 
 	kobject_init(&rdev->kobj, &rdev_ktype);
 
-	size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
+	size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
 	if (!size) {
 		printk(KERN_WARNING 
 			"md: %s has zero or unknown size, marking faulty!\n",
@@ -5235,7 +5235,7 @@ static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
 
 		if (!mddev->persistent) {
 			printk(KERN_INFO "md: nonpersistent superblock ...\n");
-			rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
+			rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
 		} else 
 			rdev->sb_start = calc_dev_sboffset(rdev->bdev);
 		rdev->sectors = rdev->sb_start;
@@ -5306,7 +5306,7 @@ static int hot_add_disk(mddev_t * mddev, dev_t dev)
 	if (mddev->persistent)
 		rdev->sb_start = calc_dev_sboffset(rdev->bdev);
 	else
-		rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
+		rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
 
 	rdev->sectors = rdev->sb_start;
 
Acked-by on v1 1/1
#1 lore raw
Quoted 11 lines · On Fri, 5 Nov 2010 15:27:21 -0400 Mike Snitzer…

Convert direct reads of an inode's i_size to using i_size_read().

i_size_{read,write} use a seqcount to protect reads from accessing incomple writes. Concurrent i_size_write()s require mutual exclussion to protect the seqcount that is used by i_size_{read,write}. But i_size_read() callers do not need to use additional locking.

Cc Jens Axboe <jaxboe-5c4llco8/ftWk0Htik3J/[email protected]>
Cc Neil Brown <[email protected]>

MD bits

It is of course OK to read i_size when you hold the appropriate mutex, but we don't in any of these cases.

Thanks,
NeilBrown

Quoted 186 lines
---
 block/blk-core.c              |    4 ++--
 block/compat_ioctl.c          |    4 ++--
 block/ioctl.c                 |    6 +++---
 drivers/block/drbd/drbd_int.h |    2 +-
 drivers/md/md.c               |   18 +++++++++---------
 5 files changed, 17 insertions(+), 17 deletions(-)

NOTE: I stopped short of fixing drbd and nbd to use revalidate_disk and/or i_size_write. I also have no intention of auditing all the direct i_size readers/writers in fs/, maybe others will have an interest.

block/blk-core.c +1 -1
index f0834e2..17fcb83 100644
@@ -1351,7 +1351,7 @@ static void handle_bad_sector(struct bio *bio)
 			bdevname(bio->bi_bdev, b),
 			bio->bi_rw,
 			(unsigned long long)bio->bi_sector +
 bio_sectors(bio),
-			(long long)(bio->bi_bdev->bd_inode->i_size
 >> 9));
+			(long
 long)(i_size_read(bio->bi_bdev->bd_inode) >> 9)); 
block/blk-core.c +1 -1
 	set_bit(BIO_EOF, &bio->bi_flags);
 }
@@ -1404,7 +1404,7 @@ static inline int bio_check_eod(struct bio
 *bio, unsigned int nr_sectors) return 0;
 
 	/* Test device or partition size, when known. */
-	maxsector = bio->bi_bdev->bd_inode->i_size >> 9;
+	maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
 	if (maxsector) {
 		sector_t sector = bio->bi_sector;
 
block/compat_ioctl.c +0 -0
index 119f07b..58c6ee5 100644
@@ -744,13 +744,13 @@ long compat_blkdev_ioctl(struct file *file,

unsigned cmd, unsigned long arg) bdi->ra_pages = (arg * 512) / PAGE_CACHE_SIZE; return 0;

block/compat_ioctl.c +2 -2
 	case BLKGETSIZE:
-		size = bdev->bd_inode->i_size;
+		size = i_size_read(bdev->bd_inode);
 		if ((size >> 9) > ~0UL)
 			return -EFBIG;
 		return compat_put_ulong(arg, size >> 9);
 
 	case BLKGETSIZE64_32:
-		return compat_put_u64(arg, bdev->bd_inode->i_size);
+		return compat_put_u64(arg,

i_size_read(bdev->bd_inode));

block/compat_ioctl.c +0 -0
 	case BLKTRACESETUP32:
 	case BLKTRACESTART: /* compatible */
block/ioctl.c +3 -3
index d724ceb..38aa194 100644
@@ -125,7 +125,7 @@ static int blk_ioctl_discard(struct block_device
 *bdev, uint64_t start, start >>= 9;
 	len >>= 9;
 
-	if (start + len > (bdev->bd_inode->i_size >> 9))
+	if (start + len > (i_size_read(bdev->bd_inode) >> 9))
 		return -EINVAL;
 	if (secure)
 		flags |= BLKDEV_DISCARD_SECURE;
@@ -307,12 +307,12 @@ int blkdev_ioctl(struct block_device *bdev,
 fmode_t mode, unsigned cmd, ret = blkdev_reread_part(bdev);
 		break;
 	case BLKGETSIZE:
-		size = bdev->bd_inode->i_size;
+		size = i_size_read(bdev->bd_inode);
 		if ((size >> 9) > ~0UL)
 			return -EFBIG;
 		return put_ulong(arg, size >> 9);
 	case BLKGETSIZE64:
-		return put_u64(arg, bdev->bd_inode->i_size);
+		return put_u64(arg, i_size_read(bdev->bd_inode));
 	case BLKTRACESTART:
 	case BLKTRACESTOP:
 	case BLKTRACESETUP:
(quoted diff) +0 -0

b/drivers/block/drbd/drbd_int.h index 9bdcf43..2637f49 100644

drivers/block/drbd/drbd_int.h +0 -0
@@ -1874,7 +1874,7 @@ static inline sector_t

drbd_md_last_sector(struct drbd_backing_dev *bdev) static inline sector_t drbd_get_capacity(struct block_device *bdev) {

drivers/block/drbd/drbd_int.h +1 -1
 	/* return bdev ? get_capacity(bdev->bd_disk) : 0; */
-	return bdev ? bdev->bd_inode->i_size >> 9 : 0;
+	return bdev ? i_size_read(bdev->bd_inode) >> 9 : 0;
 }
 
 /**
drivers/md/md.c +0 -0
index 4e957f3..02607b1 100644
@@ -706,7 +706,7 @@ static struct mdk_personality *find_pers(int

level, char *clevel) /* return the offset of the super block in 512byte sectors */ static inline sector_t calc_dev_sboffset(struct block_device *bdev) {

drivers/md/md.c +1 -1
-	sector_t num_sectors = bdev->bd_inode->i_size / 512;
+	sector_t num_sectors = i_size_read(bdev->bd_inode) / 512;
 	return MD_NEW_SIZE_SECTORS(num_sectors);
 }
@@ -1386,7 +1386,7 @@ static int super_1_load(mdk_rdev_t *rdev,

mdk_rdev_t *refdev, int minor_version) */

drivers/md/md.c +2 -2
 	switch(minor_version) {
 	case 0:
-		sb_start = rdev->bdev->bd_inode->i_size >> 9;
+		sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
 		sb_start -= 8*2;
 		sb_start &= ~(sector_t)(4*2-1);
 		break;
@@ -1472,7 +1472,7 @@ static int super_1_load(mdk_rdev_t *rdev,
 mdk_rdev_t *refdev, int minor_version) ret = 0;
 	}
 	if (minor_version)
-		rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
+		rdev->sectors = (i_size_read(rdev->bdev->bd_inode)
 >> 9) - le64_to_cpu(sb->data_offset);
 	else
 		rdev->sectors = rdev->sb_start;
@@ -1680,7 +1680,7 @@ super_1_rdev_size_change(mdk_rdev_t *rdev,

sector_t num_sectors) return 0; /* component must fit device */

drivers/md/md.c +2 -4
 	if (rdev->sb_start < rdev->data_offset) {
 		/* minor versions 1 and 2; superblock before data */
-		max_sectors = rdev->bdev->bd_inode->i_size >> 9;
+		max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
 		max_sectors -= rdev->data_offset;
 		if (!num_sectors || num_sectors > max_sectors)
 			num_sectors = max_sectors;
@@ -1690,7 +1690,7 @@ super_1_rdev_size_change(mdk_rdev_t *rdev,
 sector_t num_sectors) } else {
 		/* minor version 0; superblock after data */
 		sector_t sb_start;
-		sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
+		sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9)
- 8*2; sb_start &= ~(sector_t)(4*2 - 1);
 		max_sectors = rdev->sectors + sb_start -
 rdev->sb_start; if (!num_sectors || num_sectors > max_sectors)
@@ -2584,7 +2584,7 @@ rdev_size_store(mdk_rdev_t *rdev, const char
 *buf, size_t len) if (!sectors)
 				return -EBUSY;
 		} else if (!sectors)
-			sectors = (rdev->bdev->bd_inode->i_size >>

9) -

drivers/md/md.c +1 -0
+			sectors = (i_size_read(rdev->bdev->bd_inode)
Quoted 1 line
Quoted 1 line

9) - rdev->data_offset;

drivers/md/md.c +0 -0
 	}
 	if (sectors < my_mddev->dev_sectors)
@@ -2797,7 +2797,7 @@ static mdk_rdev_t *md_import_device(dev_t

newdev, int super_format, int super_mi

drivers/md/md.c +1 -1
 	kobject_init(&rdev->kobj, &rdev_ktype);
 
-	size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
+	size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
 	if (!size) {
 		printk(KERN_WARNING 
 			"md: %s has zero or unknown size, marking

faulty!\n", @@ -5235,7 +5235,7 @@ static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)

drivers/md/md.c +0 -0
 		if (!mddev->persistent) {
 			printk(KERN_INFO "md: nonpersistent

superblock ...\n");

drivers/md/md.c +0 -1
-			rdev->sb_start =

rdev->bdev->bd_inode->i_size / 512;

drivers/md/md.c +1 -0
+			rdev->sb_start =

i_size_read(rdev->bdev->bd_inode) / 512; } else
rdev->sb_start =
calc_dev_sboffset(rdev->bdev); rdev->sectors = rdev->sb_start;

drivers/md/md.c +1 -1
@@ -5306,7 +5306,7 @@ static int hot_add_disk(mddev_t * mddev, dev_t
 dev) if (mddev->persistent)
 		rdev->sb_start = calc_dev_sboffset(rdev->bdev);
 	else
-		rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
+		rdev->sb_start = i_size_read(rdev->bdev->bd_inode) /
 512; 
 	rdev->sectors = rdev->sb_start;
 
maintainer Acked-by on v1 1/1
#2 lore raw
In reply to Mike Snitzer
Quoted 11 lines · Mike Snitzer

Convert direct reads of an inode's i_size to using i_size_read().

i_size_{read,write} use a seqcount to protect reads from accessing incomple writes. Concurrent i_size_write()s require mutual exclussion to protect the seqcount that is used by i_size_{read,write}. But i_size_read() callers do not need to use additional locking.

Quoted 7 lines
---
 block/blk-core.c              |    4 ++--
 block/compat_ioctl.c          |    4 ++--
 block/ioctl.c                 |    6 +++---
 drivers/block/drbd/drbd_int.h |    2 +-
 drivers/md/md.c               |   18 +++++++++---------
 5 files changed, 17 insertions(+), 17 deletions(-)

DRBD one line fix trivially

Quoted 2 lines

NOTE: I stopped short of fixing drbd and nbd to use revalidate_disk and/or i_size_write.

Ok, we'll pick up the DRBD part of it.

Signature · 5 lines
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com

DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.
#3 lore raw
In reply to Mike Snitzer
Quoted 6 lines · Mike Snitzer

Convert direct reads of an inode's i_size to using i_size_read().

i_size_{read,write} use a seqcount to protect reads from accessing incomple writes. Concurrent i_size_write()s require mutual exclussion to protect the seqcount that is used by i_size_{read,write}. But i_size_read() callers do not need to use additional locking.

Thanks Mike, I'll queue this up.

Signature · 1 line
Jens Axboe