btrfs/extent_io.c: return -ENOMEM instead of BUG_ON on allocation failure in set_extent_bit if __GFP_WAIT is set in the mask

PATCH In review Wang Sheng-Hui 2 participants 3 messages Mar 31, 2012 – Apr 5, 2012 Btrfs file system

Share & tools

Wang Sheng-Hui

author

2012-03-31 07:25 UTC #0 lore raw

PATCH 1/1 btrfs/extent_io.c: return -ENOMEM instead of BUG_ON on allocation failure in set_extent_bit if __GFP_WAIT is set in the mask

If __GFP_WAIT is set in the mask, set_extent_bit should return -ENOMEM to its caller to decide what to do on allocation failure, instead of BUG_ON.

Signed-off-by Wang Sheng-Hui <[email protected]>

---
 fs/btrfs/extent_io.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

fs/btrfs/extent_io.c +2 -1

index b67814c..fd2411e 100644
@@ -720,7 +720,8 @@ int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
 again:
 	if (!prealloc && (mask & __GFP_WAIT)) {
 		prealloc = alloc_extent_state(mask);
-		BUG_ON(!prealloc);
+		if (!prealloc)
+			return -ENOMEM;
 	}
 
 	spin_lock(&tree->lock);

Signature · 1 line

1.7.1

Jeff Mahoney

2012-04-04 13:29 UTC #1 lore raw

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Quoted 16 lines · Wang Sheng-Hui

If __GFP_WAIT is set in the mask, set_extent_bit should return -ENOMEM to its caller to decide what to do on allocation failure, instead of BUG_ON.
Signed-off-by Wang Sheng-Hui <[email protected]> ---
fs/btrfs/extent_io.c | 3 ++- 1 files changed, 2 insertions(+), 1
deletions(-)
fs/btrfs/extent_io.c index renamed from fs/btrfs/extent_io.c +0 -0
b67814c..fd2411e 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -720,7 +720,8 @@ int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,
again: if (!prealloc && (mask & __GFP_WAIT)) { prealloc =
alloc_extent_state(mask); - BUG_ON(!prealloc); + if (!prealloc) +
return -ENOMEM; }

Quoted 1 line

spin_lock(&tree->lock);

Hi Sheng-Hui -

Do you have a follow-on series to this? With the exception of the set_extent_dirty call in end_bio_extent_readpage, every set_extent_bit caller uses GFP_NOFS, which includes __GFP_WAIT. If we return a failure here, we need to catch the failure in the call sites. Otherwise, the failure will just be ignored and we'll end up with an inconsistent extent state tree.

- -Jeff

- --
Jeff Mahoney
SUSE Labs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.18 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iQIcBAEBAgAGBQJPfEzIAAoJEB57S2MheeWyqcoP/RagYePToMYOdgjjxVz2cwkf BLT5B0KWWaJa0KtEOpgQATmldmA4CCq6WmC7R/Omy1q9RD75rheKxaxuQY35NTFe czBSfVr3VeMy1A6dIK+wHOiAfheuT90xSgXJJMpdkjHmNjEA1g2nupAFltUE/L70 +xFxotBuoYuTe8u++QES0JGqnw8YI6FW0z3Z4buyoqRhgiX7TukNfXXgx4c2HI0/ cOkxyvYpDi2WF9EtESn/ekwf4Rjx9kNUX/9owmDWYgHD4MLjwSQsRbwZ7BrZEM3k aBe8yZGpgr3UuL9nGiPbWHsyBeAskvXaL6Y2YIczJSvsoP/DM+9SsHAwnqkVaBHv 8qQiQqUhF1mrdl3GBSvKJarGczrSmbWZ5X94oYq4MCKyi3SeTN/bcJS+p5EV1Tvs LMRqqj26leHnl+bYKGWIfJqQRuYcfbkzTOADbn54iMYj2aG2raUjI9D53TUIInuV Csm2hhKJn+HDwu2+gZX1F99W0ehs1ps86oGfkT4WA/H7fTmugiEYxsM1Dk71Ti7/ eadns62DYFZyzKk4/jP/8Rjygc4YM/T2F9qb+SX3hWGhIIZiz5qOxjgBZy91i+84 W+DDUHjZPU+O14nNlKFA3INGS9823ENklRkzpNUhLx7ffrCLOAS73O2rzHcaSmtk FSDWO4EqxIAi4t3LoL14
=Pxwh
-----END PGP SIGNATURE-----

Wang Sheng-Hui

author

2012-04-05 02:10 UTC #2 lore raw

Quoted 20 lines · Jeff Mahoney

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
=20
Quoted 16 lines · Wang Sheng-Hui
If __GFP_WAIT is set in the mask, set_extent_bit should return=20 -ENOMEM to its caller to decide what to do on allocation failure,=20 instead of BUG_ON.
Signed-off-by Wang Sheng-Hui <[email protected]> ---=20
fs/btrfs/extent_io.c | 3 ++- 1 files changed, 2 insertions(+), 1
deletions(-)
fs/btrfs/extent_io.c index renamed from fs/btrfs/extent_io.c +0 -0
b67814c..fd2411e 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -720,7 +720,8 @@ int set_extent_bit(struct extent_io_tree *tree, u64 start, u64 end,=20
again: if (!prealloc && (mask & __GFP_WAIT)) { prealloc =3D
alloc_extent_state(mask); - BUG_ON(!prealloc); + if (!prealloc) +
return -ENOMEM; }

Quoted 6 lines

Quoted 1 line
spin_lock(&tree->lock);
=20
Hi Sheng-Hui -
=20
Do you have a follow-on series to this? With the exception of the set_extent_dirty call in end_bio_extent_readpage, every set_extent_bi=

Quoted 7 lines

caller uses GFP_NOFS, which includes __GFP_WAIT. If we return a failure here, we need to catch the failure in the call sites. Otherwise, the failure will just be ignored and we'll end up with an inconsistent extent state tree.
=20
- -Jeff
=20

Hi Jeff,

I'll review the code and provide a list of patches later. Currently we can skip it, I think.

Thanks,
Sheng-Hui

Quoted 21 lines

- --=20
Jeff Mahoney
SUSE Labs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.18 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
=20 iQIcBAEBAgAGBQJPfEzIAAoJEB57S2MheeWyqcoP/RagYePToMYOdgjjxVz2cwkf BLT5B0KWWaJa0KtEOpgQATmldmA4CCq6WmC7R/Omy1q9RD75rheKxaxuQY35NTFe czBSfVr3VeMy1A6dIK+wHOiAfheuT90xSgXJJMpdkjHmNjEA1g2nupAFltUE/L70 +xFxotBuoYuTe8u++QES0JGqnw8YI6FW0z3Z4buyoqRhgiX7TukNfXXgx4c2HI0/ cOkxyvYpDi2WF9EtESn/ekwf4Rjx9kNUX/9owmDWYgHD4MLjwSQsRbwZ7BrZEM3k aBe8yZGpgr3UuL9nGiPbWHsyBeAskvXaL6Y2YIczJSvsoP/DM+9SsHAwnqkVaBHv 8qQiQqUhF1mrdl3GBSvKJarGczrSmbWZ5X94oYq4MCKyi3SeTN/bcJS+p5EV1Tvs LMRqqj26leHnl+bYKGWIfJqQRuYcfbkzTOADbn54iMYj2aG2raUjI9D53TUIInuV Csm2hhKJn+HDwu2+gZX1F99W0ehs1ps86oGfkT4WA/H7fTmugiEYxsM1Dk71Ti7/ eadns62DYFZyzKk4/jP/8Rjygc4YM/T2F9qb+SX3hWGhIIZiz5qOxjgBZy91i+84 W+DDUHjZPU+O14nNlKFA3INGS9823ENklRkzpNUhLx7ffrCLOAS73O2rzHcaSmtk FSDWO4EqxIAi4t3LoL14
=3DPxwh
-----END PGP SIGNATURE-----