I was kind of expecting the transaction, &tp, to be passed into xfs_trans_bmap_add_attrfork(). (And then xfs_trans_bmap_add_attrfork() which calls xfs_bmap_finish() which calls xfs_trans_dup() and so
No you're right, it took me a while to remember you're working out of git. I haven't really re-read it properly yet but thanks for this one. Cheers, -- Niv Sardi
I was kind of expecting the transaction, &tp, to be passed into xfs_trans_bmap_add_attrfork(). (And then xfs_trans_bmap_add_attrfork() which calls xfs_bmap_finish() which calls xfs_trans_dup() and so
No you're right, it took me a while to remember you're working out of git. I haven't really re-read it properly yet but thanks for this one. Cheers, -- Niv Sardi
That takes a transaction and doesn't require everything to be locked anymore. this uses the roll trans call. Signed-off-by: Niv Sardi <xaiki@xxxxxxx> -- fs/xfs/xfs_bmap.c | 79 +++++++++++++++++++++++
Care to add this below xfs_trans_bmap_add_attrfork? Also please us struct xfs_inode instead of the typedef. I think it would be much cleaner if all the transaction setup, joining etc was done in xfs
Updated patch attached, Moved case where there is no transaction back into xfs_bmap_add_attrfork() and rely on caller to call xfs_trans_roll(), [...] Done, [...] I think I got it right, but error han
That takes a transaction and doesn't require everything to be locked anymore. This doesn't commit the transaction ! so direct callers, willing to use xfs_trans_roll() should do it themselves. Change
That takes a transaction and doesn't require everything to be locked anymore. this uses the roll trans call. Signed-off-by: Niv Sardi <xaiki@xxxxxxx> -- fs/xfs/xfs_bmap.c | 79 +++++++++++++++++++++++
Care to add this below xfs_trans_bmap_add_attrfork? Also please us struct xfs_inode instead of the typedef. I think it would be much cleaner if all the transaction setup, joining etc was done in xfs
Updated patch attached, Moved case where there is no transaction back into xfs_bmap_add_attrfork() and rely on caller to call xfs_trans_roll(), [...] Done, [...] I think I got it right, but error han
That takes a transaction and doesn't require everything to be locked anymore. This doesn't commit the transaction ! so direct callers, willing to use xfs_trans_roll() should do it themselves. Change
I was kind of expecting the transaction, &tp, to be passed into xfs_trans_bmap_add_attrfork(). (And then xfs_trans_bmap_add_attrfork() which calls xfs_bmap_finish() which calls xfs_trans_dup() and so
No you're right, it took me a while to remember you're working out of git. I haven't really re-read it properly yet but thanks for this one. Cheers, -- Niv Sardi
That takes a transaction and doesn't require everything to be locked anymore. this uses the roll trans call. Signed-off-by: Niv Sardi <xaiki@xxxxxxx> -- fs/xfs/xfs_bmap.c | 79 +++++++++++++++++++++++
Care to add this below xfs_trans_bmap_add_attrfork? Also please us struct xfs_inode instead of the typedef. I think it would be much cleaner if all the transaction setup, joining etc was done in xfs
Updated patch attached, Moved case where there is no transaction back into xfs_bmap_add_attrfork() and rely on caller to call xfs_trans_roll(), [...] Done, [...] I think I got it right, but error han
That takes a transaction and doesn't require everything to be locked anymore. This doesn't commit the transaction ! so direct callers, willing to use xfs_trans_roll() should do it themselves. Change