On 07/17/2017 09:20 PM, Dave Chinner wrote: > On Mon, Jul 17, 2017 at 03:51:50PM -0500, Eric Sandeen wrote: >> >> >> On 06/28/2017 11:01 PM, Eric Sandeen wrote: >>>> I guess there's also the problem that if inodesize != 512 then what are >>>> we targeting, anyway? If inodesize = 256 then we can only hit >>>> even-numbered inodes (not so bad) but if inodesize > 512 then do we jump >>>> back to wherever the inode starts? Or just give the user what they >>>> asked for, even if it's garbage? >>> Oh, hum. Right, big inodes.... >>> >>> So I'm trying to go from disk location to inode number, but the current >>> disk location may not even be the start of an inode. Hell. Well, I'm >>> sure I could craft a test to disallow "type inode" if the current location >>> cannot be on an inode boundary. But is that too clever? Should the >>> debug tool just do what the user asks? >>> >>>> (FWIW I was fine with xfs_db being dumb and giving you exactly what you >>>> point it at, even if that makes no sense. :P) >>> yeah, if you do "daddr 0" "type agf" it'll squawk, too, so it's fine to >>> do the dumb user's bidding... >> >> So as I have it today, issuing "type inode" will actually re-align you >> to an inode offset even if you're not on one: >> >> xfs_db> sb 0 >> xfs_db> addr rootino >> xfs_db> daddr >> current daddr is 128 >> xfs_db> daddr 129 >> xfs_db> type inode >> xfs_db> daddr >> current daddr is 128 >> xfs_db> >> >> good? bad? indifferent? I'm thinking "bad" but we have no mechanism to >> read a batch of inodes in xfs_db other than by passing in a legit inode number. >> >> And we have no mechanism to read only a /single/ inode... >> >> Maybe print a warning about the re-alignment and carry on? > > xfs_db is meant to be a developer tool, not idiot-proof. :P > > i.e. some knowledge of the layout of the metadata you are trying > to examine is expected. I would not like xfs_db to change the daddr > I've specified simply because I tried to parse it as the wrong > object type.... That would be my first preference as well, but all the infrastructure we have for gathering up an inode from a cluster is predicated on being given an inode nr, not a daddr. I could hack the heck out of it and refactor it to allow clusters starting on arbitrary sectors, I guess, but I'm not sure it's worth it. For now it seems more useful than not to be able to switch from a daddr to an inode (which possibly /contains/ rather than starts at that daddr ...) -Eric > Cheers, > > Dave. > -- To unsubscribe from this list: send the line "unsubscribe linux-xfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
