Bug#663540: finds MSDOS when there is none
Package: os-prober Version: 1.49 os-prober finds some MSDOS partition on my laptop: # os-prober /dev/sda2:MS-DOS 5.x/6.x/Win3.1:MS-DOS:chain # blkid | sort /dev/sda1: UUID="063e2d53-7007-4112-822f-edcba00d2928" TYPE="ext4" LABEL="root" /dev/sda2: LABEL="debian" UUID="af0b76b9-e91e-4bbd-a64c-b1a6199982bf" TYPE="ext4" /dev/sda3: LABEL="swap" UUID="55a05f9b-346d-46c8-87a6-b0b8b2f5ad37" TYPE="swap" /dev/sda4: LABEL="export" UUID="e1f404ed-b77d-4c24-91d1-7ff69540fdfa" TYPE="ext4" # parted /dev/sda u s p Model: ATA ST9160823AS (scsi) Disk /dev/sda: 312581808s Sector size (logical/physical): 512B/512B Partition Table: msdos Number Start End SizeType File system Flags 1 2048s 62492671s 62490624s primary ext4 2 62492672s 96047103s 33554432s primary ext4 3 96047104s 104435711s 8388608sprimary linux-swap(v1) 4 104435712s 312581807s 208146096s primary ext4 This makes /dev/sda2 unbootable. At boot time grub2 gives me error: invalid signature. os-prober 1.42 from Stable doesn't show this problem: # os-prober /dev/sda2:Debian GNU/Linux (wheezy/sid):Debian:linux Regards Harri -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#663540: finds MSDOS when there is none
On 12.03.2012 08:57, Harald Dunkel wrote: > Package: os-prober > Version: 1.49 > > os-prober finds some MSDOS partition on my laptop: [os-prober says /dev/sda2 is MSDOS, while parted and blkid identifies it as ext4] I understand that, due to the fact that other utils correctly identifies this filesystem as ext4, there's a bug in os-prober which can be corrected. But anyway, this is a very minor situation which can be corrected on your side once and forever. I've no idea about level of your system and technology knowlege, so I assume a beginner level, -- please forgive me if you're not a beginner :) All utils examine first few sectors of a filesystem to determine its type based on some heuristics. Ext* family (ext2, ext3, ext4) on purpose never, ever, touch first 512 bytes of the filesystem, these bytes are unused. This place is reserved for a bootloader. So in order to detect ext* filesystem, any tool should inspect subsequent sectors. But MSDOS (FAT) filesystem, on the other hand, does use first 512 bytes and puts its "signature" there. So a more naive approach at determining the filesystem type is to check first 512 bytes, and if MS-DOS signature is found there, report that it is MS-DOS. This is, apparently, what os-prober is currently doing. I suspect you actually had ms-dos filesystem in this partition before, but later installed linux and ext4 filesystem on it. If this is the case, just zero out the first 512 bytes and no tool, ever, even the ones who does not know anything about ext*, will detect it as ms-dos filesystem. How: dd if=/dev/zero bs=512 count=1 of=/dev/sda2 This will zero-out first 512 bytes on your sda2. If you're unsure, you may save it before doing so, to a file on _another_ filesystem, eg sda1 (root fs), like this: dd if=/dev/sda2 bs=512 count=1 of=/sda2-beginning.bin It will create file named /sda2-beginning.bin, of length 512 bytes, and will save the start of your sda2 in it. You can restore it back by doing dd if=/sda2-beginning.bin of=/dev/sda2 (or, in case you booted a rescue CD or something like that, mount your regular root fs and specify corresponding path there). This may sound/smell/etc very dangerous, but if it really _is_ ext4 (it appears to be) and if it is not a boot partition (it is not), it is safe thing to do. Thanks, /mjt -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#663540: finds MSDOS when there is none
Michael Tokarev wrote: > So a more naive approach at determining the filesystem type is to > check first 512 bytes, and if MS-DOS signature is found there, > report that it is MS-DOS. This is, apparently, what os-prober is > currently doing. os-prober does not look at fragile magic numbers in filsystem headers. It mounts the actual filesystem, and looks for (fragile) specific filenames indicating which OS is contained in it. The particular OS string we see here comes from: # MS-DOS if [ -z "$found" ] && item_in_dir -q dos "$2" && [ -d "$2"/"$(item_in_dir dos "$2")" ]; then long="MS-DOS 5.x/6.x/Win3.1" short=MS-DOS found=true fi Perhaps your filesystem has a "/dos" directory? item_in_dir checks for any case, so /DOS or /Dos would also do. These checks don't try to avoid false positives like a /dos or /windows, because the code earlier checks that the mounted filesystem is one of ntfs, vfat, or msdos. To which list was semi-recently added fuse, and I suspect that might be the root of the problem. Colin, if grub-mount is being used, will *every* filesystem appear as fuse? If so, this and other code has become much more prone to false positives than it was when it was originally written. (os-prober 1.42 didn't have this problem I think only because the above code in it had a typo that made it never match anything!) -- see shy jo signature.asc Description: Digital signature
Bug#663540: finds MSDOS when there is none
Joey Hess, le Mon 12 Mar 2012 11:14:07 -0400, a écrit : > The particular OS string we see here comes from: > > # MS-DOS > if [ -z "$found" ] && item_in_dir -q dos "$2" && > [ -d "$2"/"$(item_in_dir dos "$2")" ]; then > long="MS-DOS 5.x/6.x/Win3.1" > short=MS-DOS > > found=true > fi > > item_in_dir checks for any case, so /DOS or /Dos would also do. > Perhaps your filesystem has a "/dos" directory? > > These checks don't try to avoid false positives like a /dos or /windows, > because the code earlier checks that the mounted filesystem is one of ntfs, > vfat, or msdos. Mmm, why not looking for io.sys at the root? In all the versions of DOS that I know there has to be one for the partition to be bootable. Samuel -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#663540: finds MSDOS when there is none
On Mon, Mar 12, 2012 at 11:14:07AM -0400, Joey Hess wrote: > These checks don't try to avoid false positives like a /dos or /windows, > because the code earlier checks that the mounted filesystem is one of ntfs, > vfat, or msdos. To which list was semi-recently added fuse, and I suspect > that might be the root of the problem. > > Colin, if grub-mount is being used, will *every* filesystem appear as fuse? > If so, this and other code has become much more prone to false positives > than it was when it was originally written. Hm, probably, yes. (To some extent, you get to pick which kind of fragility you want.) How about, if grub-mount is used, we use grub-probe to find out which GRUB filesystem driver is in use, and stash that somewhere so that individual tests can get at it? That ought to be just as reliable as the existing OS filesystem type checks. -- Colin Watson [cjwat...@debian.org] -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#663540: finds MSDOS when there is none
Samuel Thibault wrote: > Mmm, why not looking for io.sys at the root? In all the versions of DOS > that I know there has to be one for the partition to be bootable. According to http://en.wikipedia.org/wiki/IO.SYS , some of the DOS clones did not use io.sys, which may be why I don't remember it well. I added a check for autoexec.bat || config.sys , just to have an added check and avoid the worse of the false positives. My gut feeling is that either a) nobody cares if their DOS is not detected, or b) if they do care, they have one of those files. -- see shy jo signature.asc Description: Digital signature
Bug#663540: finds MSDOS when there is none
On Mon, Mar 12, 2012 at 12:11:29PM -0400, Joey Hess wrote: > Colin Watson wrote: > > How about, if grub-mount is used, we use grub-probe to find out which > > GRUB filesystem driver is in use, and stash that somewhere so that > > individual tests can get at it? That ought to be just as reliable as > > the existing OS filesystem type checks. > > Yes please. I've CCed the other bug I opened, since it appears there > are also several false negatives when grub-probe is used. Your method > should solve it. > > Once we have a real FS type from grub-probe, the best thing would be to > pass it into individual checks as $3 in place of "fuse". Thanks. I've done that, and I think I've sorted out all the examples you pointed out as a result. qnx4 doesn't matter since AFAIK GRUB can't mount its filesystem, so grub-mount will fail and we'll fall back to the OS driver. If grub-mount ever learns about it and calls its driver something other than "qnx4" then we can add that. -- Colin Watson [cjwat...@debian.org] -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#663540: finds MSDOS when there is none
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 03/12/12 17:05, Joey Hess wrote: > > I added a check for autoexec.bat || config.sys , just to have an added check > and avoid the worse of the false positives. My gut feeling is that either a) > nobody cares if their DOS is not detected, or b) if they do care, they have > one of those files. > ??? I understand that os-prober is some kind of heuristic, but wouldn't it be more wise to delay the look at the first sector until the file system has been examined? Obviously the information stored in sector #0 is less reliable. Anyway, many thanx for the fix. I just tried it on my laptop. Seems to work. Regards Harri -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) iEYEARECAAYFAk9jrLUACgkQUTlbRTxpHjdtEwCfbiPVBbtOmKjT9lU45V0ilGF6 ekIAnA5XlwUYG26SLLA5YyoG1q0Uucvg =LusM -END PGP SIGNATURE- -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
Bug#663600: Bug#663540: finds MSDOS when there is none
Colin Watson wrote: > How about, if grub-mount is used, we use grub-probe to find out which > GRUB filesystem driver is in use, and stash that somewhere so that > individual tests can get at it? That ought to be just as reliable as > the existing OS filesystem type checks. Yes please. I've CCed the other bug I opened, since it appears there are also several false negatives when grub-probe is used. Your method should solve it. Once we have a real FS type from grub-probe, the best thing would be to pass it into individual checks as $3 in place of "fuse". -- see shy jo signature.asc Description: Digital signature