sfdisk doesnt understand GUID Partition Table (GPT) and it is not designed for large partitions. In particular case use more advanced GNU parted(8).
BE EXTREMELY CAREFUL - ONE TYPING MISTAKE AND ALL YOUR DATA IS LOST
As a precaution, one can save the sectors changed by sfdisk:
% sfdisk /dev/hdd -O hdd-partition-sectors.save ... %
Then, if you discover that you did something stupid before anything else has been written to disk, it may be possible to recover the old situation with
% sfdisk /dev/hdd -I hdd-partition-sectors.save %
(This is not the same as saving the old partition table: a readable version of the old partition table can be saved using the -d option. However, if you create logical partitions, the sectors describing them are located somewhere on disk, possibly on sectors that were not part of the partition table before. Thus, the information the -O option saves is not a binary version of the output of -d.)
There are many options.
|-v or --version|
|Print version number of sfdisk and exit immediately.|
|-? or --help|
|Print a usage message and exit immediately.|
|-T or --list-types|
|Print the recognized types (system Ids).|
|-s or --show-size|
|List the size of a partition.|
|-g or --show-geometry|
|List the kernels idea of the geometry of the indicated disk(s).|
|-G or --show-pt-geometry|
|List the geometry of the indicated disks guessed by looking at the partition table.|
|-l or --list|
|List the partitions of a device.|
Dump the partitions of a device in a format useful as input
to sfdisk. For example,
|-V or --verify|
|Test whether partitions seem correct. (See above.)|
|-i or --increment|
|Number cylinders etc. starting from 1 instead of 0.|
Change only the single partition indicated. For example:
|Make the indicated partition(s) active, and all others inactive.|
|-c or --id number [Id]|
If no Id argument given: print the partition Id of the indicated
partition. If an Id argument is present: change the type (Id) of
the indicated partition to the given value.
This option has the two very long forms --print-id and --change-id.
|-uS or -uB or -uC or -uM|
|Accept or report in units of sectors (blocks, cylinders, megabytes, respectively). The default is cylinders, at least when the geometry is known.|
|-x or --show-extended|
|Also list non-primary extended partitions on output, and expect descriptors for them on input.|
|Specify the number of cylinders, possibly overriding what the kernel thinks.|
|Specify the number of heads, possibly overriding what the kernel thinks.|
|Specify the number of sectors, possibly overriding what the kernel thinks.|
|-f or --force|
|Do what I say, even if it is stupid.|
|-q or --quiet|
|Suppress warning messages.|
|-L or --Linux|
|Do not complain about things irrelevant for Linux.|
|-D or --DOS|
|For DOS-compatibility: waste a little space. (More precisely: if a partition cannot contain sector 0, e.g. because that is the MBR of the device, or contains the partition table of an extended partition, then sfdisk would make it start the next sector. However, when this option is given it skips to the start of the next track, wasting for example 33 sectors (in case of 34 sectors/track), just like certain versions of DOS do.) Certain Disk Managers and boot loaders (such as OSBS, but not LILO or the OS/2 Boot Manager) also live in this empty space, so maybe you want this option if you use one.|
|-E or --DOS-extended|
|Take the starting sector numbers of "inner" extended partitions to be relative to the starting cylinder boundary of the outer one, (like some versions of DOS do) rather than to the starting sector (like Linux does). (The fact that there is a difference here means that one should always let extended partitions start at cylinder boundaries if DOS and Linux should interpret the partition table in the same way. Of course one can only know where cylinder boundaries are when one knows what geometry DOS will use for this disk.)|
|--IBM or --leave-last|
|Certain IBM diagnostic programs assume that they can use the last cylinder on a disk for disk-testing purposes. If you think you might ever run such programs, use this option to tell sfdisk that it should not allocate the last cylinder. Sometimes the last cylinder contains a bad sector table.|
|-n||Go through all the motions, but do not actually write to disk.|
|-R||Only execute the BLKRRPART ioctl (to make the kernel re-read the partition table). This can be useful for checking in advance that the final BLKRRPART will be successful, and also when you changed the partition table by hand (e.g., using dd from a backup). If the kernel complains (device busy for revalidation (usage = 2)) then something still uses the device, and you still have to unmount some file system, or say swapoff to some swap partition.|
|When starting a repartitioning of a disk, sfdisk checks that this disk is not mounted, or in use as a swap device, and refuses to continue if it is. This option suppresses the test. (On the other hand, the -f option would force sfdisk to continue even when this test fails.)|
|-O file||Just before writing the new partition, output the sectors that are going to be overwritten to file (where hopefully file resides on another disk, or on a floppy).|
After destroying your filesystems with an unfortunate
sfdisk command, you would have been able to restore the old situation
if only you had preserved it using the -O flag.
A partition descriptor has 6 fields:
The two hsc fields indicate head, sector and cylinder of the begin and the end of the partition. Since each hsc field only takes 3 bytes, only 24 bits are available, which does not suffice for big disks (say > 8GB). In fact, due to the wasteful representation (that uses a byte for the number of heads, which is typically 16), problems already start with 0.5GB. However Linux does not use these fields, and problems can arise only at boot time, before Linux has been started. For more details, see the lilo documentation.
Each partition has a type, its Id, and if this type is 5 or f (extended partition) the starting sector of the partition again contains 4 partition descriptors. MSDOS only uses the first two of these: the first one an actual data partition, and the second one again an extended partition (or empty). In this way one gets a chain of extended partitions. Other operating systems have slightly different conventions. Linux also accepts type 85 as equivalent to 5 and f - this can be useful if one wants to have extended partitions under Linux past the 1024 cylinder boundary, without DOS FDISK hanging. (If there is no good reason, you should just use 5, which is understood by other systems.)
Partitions that are not primary or extended are called logical. Often, one cannot boot from logical partitions (because the process of finding them is more involved than just looking at the MBR). Note that of an extended partition only the Id and the start are used. There are various conventions about what to write in the other fields. One should not try to use extended partitions for data storage or swap.
Fields are separated by whitespace, or comma or semicolon possibly followed by whitespace; initial and trailing whitespace is ignored. Numbers can be octal, decimal or hexadecimal, decimal is default. When a field is absent or empty, a default value is used.
The <c,h,s> parts can (and probably should) be omitted - sfdisk computes them from <start> and <size> and the disk geometry as given by the kernel or specified using the -H, -S, -C flags.
Bootable is specified as [*|-], with as default not-bootable. (The value of this field is irrelevant for Linux - when Linux runs it has been booted already - but might play a role for certain boot loaders and for other operating systems. For example, when there are several primary DOS partitions, DOS assigns C: to the first among these that is bootable.)
Id is given in hex, without the 0x prefix, or is [E|S|L|X], where L (LINUX_NATIVE (83)) is the default, S is LINUX_SWAP (82), E is EXTENDED_PARTITION (5), and X is LINUX_EXTENDED (85).
The default value of start is the first nonassigned sector/cylinder/...
The default value of size is as much as possible (until next partition or end-of-disk).
However, for the four partitions inside an extended partition, the defaults are: Linux partition, Extended partition, Empty, Empty.
But when the -N option (change a single partition only) is given, the default for each field is its previous value.
sfdisk /dev/hdc << EOF 0,407 ,407 ; ; EOF
sfdisk /dev/hdb << EOF ,3,L ,60,L ,19,S ,,E ,130,L ,130,L ,130,L ,,L EOF
With the -x option, the number of input lines must be a multiple of 4: you have to list the two empty partitions that you never want using two blank lines. Without the -x option, you give one line for the partitions inside a extended partition, instead of four, and terminate with end-of-file (^D). (And sfdisk will assume that your input line represents the first of four, that the second one is extended, and the 3rd and 4th are empty.)
The DOS 6.x FORMAT command looks for some information in the first sector of the data area of the partition, and treats this information as more reliable than the information in the partition table. DOS FORMAT expects DOS FDISK to clear the first 512 bytes of the data area of a partition whenever a size change occurs. DOS FORMAT will look at this extra information even if the /U flag is given -- we consider this a bug in DOS FORMAT and DOS FDISK.
The bottom line is that if you use sfdisk to change the size of a DOS partition table entry, then you must also use dd to zero the first 512 bytes of that partition before using DOS FORMAT to format the partition. For example, if you were using sfdisk to make a DOS partition table entry for /dev/hda1, then (after exiting sfdisk and rebooting Linux so that the partition table information is valid) you would use the command "dd if=/dev/zero of=/dev/hda1 bs=512 count=1" to zero the first 512 bytes of the partition. BE EXTREMELY CAREFUL if you use the dd command, since a small typo can make all of the data on your disk useless.
For best results, you should always use an OS-specific partition table program. For example, you should make DOS partitions with the DOS FDISK program and Linux partitions with the Linux sfdisk program.
Stephen Tweedie reported (930515): Most reports of superblock corruption turn out to be due to bad partitioning, with one filesystem overrunning the start of the next and corrupting its superblock. I have even had this problem with the supposedly-reliable DRDOS. This was quite possibly due to DRDOS-6.0s FDISK command. Unless I created a blank track or cylinder between the DRDOS partition and the immediately following one, DRDOS would happily stamp all over the start of the next partition. Mind you, as long as I keep a little free disk space after any DRDOS partition, I dont have any other problems with the two coexisting on the one drive.
A. V. Le Blanc writes in README.efdisk: Dr. DOS 5.0 and 6.0 has been reported to have problems cooperating with Linux, and with this version of efdisk in particular. This efdisk sets the system type to hexadecimal 81. Dr. DOS seems to confuse this with hexadecimal 1, a DOS code. If you use Dr. DOS, use the efdisk command t to change the system code of any Linux partitions to some number less than hexadecimal 80; I suggest 41 and 42 for the moment.
A. V. Le Blanc writes in his README.fdisk: DR-DOS 5.0 and 6.0 are reported to have difficulties with partition ID codes of 80 or more. The Linux fdisk used to set the system type of new partitions to hexadecimal 81. DR-DOS seems to confuse this with hexadecimal 1, a DOS code. The values 82 for swap and 83 for file systems should not cause problems with DR-DOS. If they do, you may use the fdisk command t to change the system code of any Linux partitions to some number less than hexadecimal 80; I suggest 42 and 43 for the moment.
In fact, it seems that only 4 bits are significant for the DRDOS FDISK, so that for example 11 and 21 are listed as DOS 2.0. However, DRDOS itself seems to use the full byte. I have not been able to reproduce any corruption with DRDOS or its fdisk.
There are too many options.
There is no support for non-DOS partition types.