sg_format - format or resize a SCSI disk (perhaps change its block size)

sg_format [--count=<n>] [--early] [--format] [--help] [--long] [--resize] [--size=<n>] [--verbose] [--version] [--wait] <scsi_device>

Not all SCSI direct access devices need to be formatted and some have vendor specific formatting procedures. SCSI disks with rotating media are probably the largest group that do support a 'standard' format operation. They are typically factory formatted to a block size of 512 bytes with the largest number of blocks that the manufacturer recommends. That number of blocks typically leaves aside a certain number of tracks and sectors for reassignment of logical block addresses during the life of the disk.

This utility can format modern SCSI disks and potentially change their block size (if permitted) and the block count (i.e. number of accessible blocks on the media also known as "resizing"). Resizing a disk to less than the manufacturer's recommended block count is sometimes called "short stroking" (see NOTES section). Resizing the block count while not changing the block size may not require a format operation. Recent changes to the SBC-2 draft standard (see www.t10.org) have obsoleted the "format device" mode page. Many of the low level details found in that mode page are now left up to the discretion of the manufacturer.

When this utility is used without options (apart from a device name) then it prints out the existing block size and block count derived from two sources. These two sources are a block descriptor in the response to a MODE SENSE command and the response to a READ CAPACITY command. The reason for this double check is to detect a "format corrupt" state (see NOTES section).

A recent addition in the SBC-2 is "protection information". See the section of that name (section 4.16 in draft SBC-2 rev 16). It adds an extra 8 bytes of protection information to each block (a 2 byte "logical block guard" (CRC), a 2 byte "logical block application guard", and a 4 byte "logical block reference tag"). A device that supports protection information sets the "protect" bit in a standard INQUIRY response. The "FMTPINFO" and "RTO_REQ" bits in the FORMAT command cdb are associated with protection information and can be set by this utility.

--count=<n> | -c <n>

count of blocks to be formatted or media to be resized to. Can be used with either "--format" or "--resize". With "--format" this option need not be given in which case it is assumed to be zero. With "--format" the interpretation of <n> is:

(n > 0) : only format the first <n> blocks and READ

CAPACITY will report <n> blocks after format

(n = 0) and block size unchanged : use existing block count

(n = 0) and block size changed : recommended maximum block

count for new block size

(n = -1) : use recommended maximum block count

(n < -1) : illegal

With "--resize" this option must be given and <n> has this interpretation:

(n > 0) : after resize READ CAPACITY will report <n>

blocks

(n = 0) : after resize READ CAPACITY will report 0 blocks

(n = -1) : after resize READ CAPACITY will report its

maximum number of blocks

(n < -1) : illegal

In both cases if the given <n> exceeds the maximum number of blocks (for the block size) then the disk reports an error. See NOTES section below.

--early | -e

this option is active when "--format" is given. The default action of this utility is to poll the disk every 30 seconds to determine the progress of the format operation until it is finished. When this option is given this utility will exit "early" as soon as the format has commenced. Then the user can monitor the progress of the ongoing format operation with other utilities (e.g. sg_turs or sg_requests). This option and "--wait" cannot both be given.

--format | -F

issue a SCSI FORMAT command. This will destroy all the data held on the media. This option is required to change the block size of a disk. See NOTES section for implementation details and EXAMPLES section for typical use.

--help | -h

print out the usage information then exit.

--long | -l

the default action of this utility is to assume 32 bit logical block addresses. With 512 byte block size this permits almost 2 terabytes (almost 2 ** 41 bytes) on a single disk. This option selects commands and parameters that allow for 64 bit logical block addresses. Specifically this is the "longlba" flag in the MODE SENSE (10) command and READ CAPACITY (16) rather than READ CAPACITY (10). When a disk supports "protection information" then this option may also be useful.

--pinfo | -p

instructs a '--format' to add an extra 8 bytes of protection information. Default action is not to format with protection information. Has no action unless '--format' is given.

--resize | -r

rather than format the disk, it can be resized. This means changing the number of blocks on the device reported by the READ CAPACITY command. This option should be used with the '--count=<new_blk_count>' option. The contents of all logical blocks on the media remain unchanged when this option is used. This means that any resize operation can be reversed. This option cannot be used together with either "--format" or a "--size" whose argument is different to the existing block size.

--rto_req | -R

instructs a format to enable application client ownership of the "logical block reference tag" field. The default action is to disable application client ownership of that field. Has no action unless both '--format' and '--pinfo' are given.

--size=<n> | -s <n>

block size (i.e. number of bytes in each block) to format the device to. The default value is whatever is currently reported by the block descriptor in a MODE SENSE command. This option is only active when the "--format" option is also given. If the block size given by this option is different from the current value then a MODE SELECT command is used to change it prior to the FORMAT command being started (as recommended in the draft standard). Recent SCSI disks usually have 512 byte sectors by default and allow up to 16 bytes extra in a sector (i.e. 528 byte sectors). If the given size in unacceptable to the disk, most likely an "Invalid field in parameter list" message will appear in sense data (requires the use of '-v' to decode sense data).

--verbose | -v

increase the level of verbosity, (i.e. debug output). "-vvv" gives the maximum debug output.

--version | -V

print the version string and then exit.

--wait | -w

this option only has an effect when used together with the "--format" option. The default format action is to set the "IMMED" bit in the FORMAT UNIT command's (short) parameter header. If this option (i.e. "--wait") is given then the "IMMED" bit is not set. Then the FORMAT UNIT command waits until the format operation completes before returning its response. This can be several hours on large disks. This utility sets a four hour timeout on such a FORMAT UNIT command.

The format command is still quite complicated and a large set of variables in the format command itself and associated parameters are set to default values by this utility. These include: LONGLIST, CMPLIST and the defect list format (in the command) and all flags within the parameter header apart from IMMED (as explained in the "--wait" option).

The SBC-2 draft standard (revision 16) says that the REQUEST SENSE command should be used for obtaining a progress indication when the format command returns prior to the completion of the format operation. However, tests on a selection of recent disks shows that TEST UNIT READY commands yield progress indications (but not REQUEST SENSE commands). A new option may be required to handle this when disks catch up to the current draft.

When the "--format" option is given then there is a 10 second window during which the user is invited to abort sg_format. This is just prior the FORMAT UNIT SCSI command being issued. If the "--wait" option is not given then the FORMAT UNIT SCSI command is issued with the IMMED bit set which causes the SCSI command to return after it has started the format operation. The "--early" option will cause sg_format to exit at that point. Otherwise the given device is polled every 30 seconds with TEST UNIT READY commands until it reports an "all clear" (i.e. the format operation has completed). Normally these polling commands will result in a progress indicator (expressed as a percentage) being output to the screen. If the user gets bored watching the progress report then sg_format can be terminated (e.g. with control-C) without effecting the format operation which continues. However a bus or device reset (or a power cycle) may well cause the device to become "format corrupt".

When the "--format" and "--wait" options are both given then this utility may take a long time to return. In this case care should be taken not to send any other SCSI commands to the disk as it may not respond leaving those commands queued behind the active format command. This may cause a timeout in the OS driver (in a lot shorter period than 4 hours applicable to the format command). This may result in the OS resetting the disk leaving the format operation incomplete. This may leave the disk in a "format corrupt" state requiring another format to remedy the situation.

When the block size (i.e. the number of bytes in each block) is changed on a disk two SCSI commands must be sent: a MODE SELECT to change the block size followed by a FORMAT command. If the MODE SELECT command succeeds and the FORMAT fails then the disk may be in a state that the draft standard calls "format corrupt". A block descriptor in a subsequent MODE SENSE will report the requested new block size while a READ CAPACITY command will report the existing (i.e. different) block size. Alternatively the READ CAPACITY command may fail, reporting the device is not ready, potentially requiring a format. The solution to this situation is to do a format again (and this time the new block size does not have to be given) or change the block size back to the original size.

The draft SBC-2 standard states that the block count can be set back to the manufacturer's maximum recommended value in a format or resize operation. This can be done by placing an address of 0xffffffff (or the 64 bit equivalent) in the appropriate block descriptor field to a MODE SELECT command. In signed (two's complement) arithmetic that value corresponds to '-1'. So a "--count" argument of '-1' causes the block count to be set back to the manufacturer's maximum recommended value. To see exactly which SCSI commands are being executed and parameters passed add "-vvv" to the sg_format command line.

Short stroking is a technique to trade off capacity for performance. Disk performance is usually highest on the outer tracks (i.e. lower logical block addresses) so by resizing or reformatting a disk to a smaller capacity, average performance will usually be increased.

Other utilities may be useful in finding information associated with formatting. These include sg_inq to fetch standard INQUIRY information (e.g. the PROTECT bit) and to fetch the extended INQUIRY VPD page (e.g. RTO and GRD_CHK bits). The sdparm (or sginfo) utility can be used to access and potentially change the now obsolete format mode page.

scsiformat is another utility available for formatting SCSI disks with linux. It dates from 1997 (most recent update) and may be useful for disks whose firmware is of that vintage.

The argument to "--count" is a number which may be followed by one of these multiplicative suffixes: c C *1; w W *2; b B *512; k K KiB *1,024; KB *1,000; m M MiB *1,048,576; MB *1,000,000 . This pattern continues for "G", "T" and "P". Also a suffix of the form "x<n>" multiplies the leading number by <n>. Alternatively numerical values can be given in hexadecimal preceded by either "0x" or "0X". When hex numbers are given, multipliers cannot be used.

First, do not alter anything but print out the existing block count and size derived from two sources: a block descriptor in a MODE SELECT command response and from the response of a READ CAPACITY commands:

sg_format /dev/sdm

Now a simple format, leaving the block count and size as they were previously, executing the format command in IMMED mode and polling every 30 seconds to print out a progress indication:

sg_format --format /dev/sdm

Now the same format, but waiting (passively) until the format operation is complete:

sg_format --format --wait /dev/sdm

Next is a format in which the block size is changed to 520 bytes and the block count is set to the manufacturer's maximum value (for that block size). Note, not all disks support changing the block size:

sg_format --format --size=520 /dev/sdm

Now a resize operation so that only the first 0x10000 (65536) blocks on a disk are accessible. The remaining blocks remain unaltered.

sg_format --resize --count=0x10000 /dev/sdm

Now resize the disk back to its normal (maximum) block count:

sg_format --resize --count=-1 /dev/sdm

Written by Grant Grundler, James Bottomley and Douglas Gilbert.

Report bugs to <dgilbert at interlog dot com>.

Copyright © 2005 Grant Grundler, James Bottomley and Douglas Gilbert

This software is distributed under the GPL version 2. There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

sg_turs, sg_requests, sg_inq, sg_modes, sginfo, sg_wr_mode (all in sg3_utils), sdparm, scsiformat, setblocksize