Skip to contentSkip to conversation

add I/O error uevent for block devices

PATCH In review Kay Sievers 2 participants 6 messages Feb 18, 2005

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 add I/O error uevent for block devices

For HAL we want to get notified about I/O errors of block devices. This is especially useful for devices we are unable to poll and therefore can't know if something goes wrong here.

===== fs/buffer.c 1.270 vs edited =====

fs/buffer.c +3 -1
@@ -105,6 +105,7 @@ static void buffer_io_error(struct buffe
 	printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n",
 			bdevname(bh->b_bdev, b),
 			(unsigned long long)bh->b_blocknr);
+	kobject_uevent(&bh->b_bdev->bd_disk->kobj, KOBJ_IO_ERROR, NULL);
 }
 
 /*
@@ -550,7 +551,8 @@ static void end_buffer_async_read(struct
 		set_buffer_uptodate(bh);
 	} else {
 		clear_buffer_uptodate(bh);
-		buffer_io_error(bh);
+		if (printk_ratelimit())
+			buffer_io_error(bh);
 		SetPageError(page);
 	}
 

===== include/linux/kobject_uevent.h 1.6 vs edited =====

include/linux/kobject_uevent.h +1 -0
@@ -29,6 +29,7 @@ enum kobject_action {
 	KOBJ_UMOUNT	= (__force kobject_action_t) 0x05,	/* umount event for block devices */
 	KOBJ_OFFLINE	= (__force kobject_action_t) 0x06,	/* offline event for hotplug devices */
 	KOBJ_ONLINE	= (__force kobject_action_t) 0x07,	/* online event for hotplug devices */
+	KOBJ_IO_ERROR	= (__force kobject_action_t) 0x08,	/* I/O error for devices */
 };
 
 

===== lib/kobject_uevent.c 1.18 vs edited =====

lib/kobject_uevent.c +2 -0
@@ -44,6 +44,8 @@ static char *action_to_string(enum kobje
 		return "offline";
 	case KOBJ_ONLINE:
 		return "online";
+	case KOBJ_IO_ERROR:
+		return "io_error";
 	default:
 		return NULL;
 	}
#1 lore raw
Quoted 8 lines · Kay Sievers

For HAL we want to get notified about I/O errors of block devices.
This is especially useful for devices we are unable to poll and therefore can't know if something goes wrong here.

Quoted 9 lines

===== fs/buffer.c 1.270 vs edited =====
--- 1.270/fs/buffer.c 2005-01-21 06:02:13 +01:00
+++ edited/fs/buffer.c 2005-02-17 22:56:05 +01:00
@@ -105,6 +105,7 @@ static void buffer_io_error(struct buffe
printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n",
bdevname(bh->b_bdev, b),
(unsigned long long)bh->b_blocknr);
+ kobject_uevent(&bh->b_bdev->bd_disk->kobj, KOBJ_IO_ERROR, NULL);
}

- buffer_io_error() is called from interrupt context, and kobject_uevent() does multiple GFP_KERNEL allocations. You'll need to use kobject_uevent_atomic().

- the prink_ratelimit() fix in end_buffer_async_read() should be a separate patch, really. I'll fix that up.

- there are numerous other places where an I/O error can be detected:
grep the tree for b_end_io and bio_end_io.

#2 lore raw
Quoted 9 lines · Andrew Morton
Quoted 4 lines · Kay Sievers

For HAL we want to get notified about I/O errors of block devices.
This is especially useful for devices we are unable to poll and therefore can't know if something goes wrong here.

- buffer_io_error() is called from interrupt context, and kobject_uevent() does multiple GFP_KERNEL allocations. You'll need to use kobject_uevent_atomic().

Fixed.

Quoted 2 lines

- the prink_ratelimit() fix in end_buffer_async_read() should be a separate patch, really. I'll fix that up.

Removed that part.

Quoted 2 lines

- there are numerous other places where an I/O error can be detected:
grep the tree for b_end_io and bio_end_io.

You mean the mmap and direct-io stuff?

Thanks,
Kay

-----
For HAL we want to get notified about I/O errors of block devices. This is especially useful for devices we are unable to poll and therefore can't know if something goes wrong here.

===== fs/buffer.c 1.270 vs edited =====

fs/buffer.c +1 -0
@@ -105,6 +105,7 @@ static void buffer_io_error(struct buffe
 	printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n",
 			bdevname(bh->b_bdev, b),
 			(unsigned long long)bh->b_blocknr);
+	kobject_uevent_atomic(&bh->b_bdev->bd_disk->kobj, KOBJ_IO_ERROR, NULL);
 }
 
 /*

===== include/linux/kobject_uevent.h 1.6 vs edited =====

include/linux/kobject_uevent.h +1 -0
@@ -29,6 +29,7 @@ enum kobject_action {
 	KOBJ_UMOUNT	= (__force kobject_action_t) 0x05,	/* umount event for block devices */
 	KOBJ_OFFLINE	= (__force kobject_action_t) 0x06,	/* offline event for hotplug devices */
 	KOBJ_ONLINE	= (__force kobject_action_t) 0x07,	/* online event for hotplug devices */
+	KOBJ_IO_ERROR	= (__force kobject_action_t) 0x08,	/* I/O error for devices */
 };
 
 

===== lib/kobject_uevent.c 1.18 vs edited =====

lib/kobject_uevent.c +2 -0
@@ -44,6 +44,8 @@ static char *action_to_string(enum kobje
 		return "offline";
 	case KOBJ_ONLINE:
 		return "online";
+	case KOBJ_IO_ERROR:
+		return "io_error";
 	default:
 		return NULL;
 	}
#3 lore raw
Quoted 5 lines · Kay Sievers
Quoted 3 lines

- there are numerous other places where an I/O error can be detected:
grep the tree for b_end_io and bio_end_io.

You mean the mmap and direct-io stuff?

direct-io, certainly. Also reiserfs, xfs, ntfs, ext3, jfs and possibly md have their own I/O completion handlers.

#4 lore raw
Quoted 9 lines · Andrew Morton
Quoted 5 lines · Kay Sievers
Quoted 3 lines

- there are numerous other places where an I/O error can be detected:
grep the tree for b_end_io and bio_end_io.

You mean the mmap and direct-io stuff?

direct-io, certainly. Also reiserfs, xfs, ntfs, ext3, jfs and possibly md have their own I/O completion handlers.

Hmm, ok. Any idea how to propagate errors like this in a saner way? Some of these places don't even log errors and spreading uevents all over the place doesn't sounds like the best idea.

Thanks,
Kay

#5 lore raw
Quoted 7 lines · Kay Sievers
Quoted 5 lines · Andrew Morton
Quoted 4 lines · Kay Sievers
Quoted 3 lines

- there are numerous other places where an I/O error can be detected:
grep the tree for b_end_io and bio_end_io.

Quoted 9 lines
Quoted 4 lines
Quoted 1 line

You mean the mmap and direct-io stuff?

direct-io, certainly. Also reiserfs, xfs, ntfs, ext3, jfs and possibly md have their own I/O completion handlers.

Hmm, ok. Any idea how to propagate errors like this in a saner way? Some of these places don't even log errors and spreading uevents all over the place doesn't sounds like the best idea.

I guess you should add some new generic function handle_block_io_error(bdev, rw, sector) in ll_rw_blk.c and do whatever you need to do inside that. Move the ratelimited printk there, too.