How do I fix SAS drives stuck at 520-byte sectors?

1. 1.List the SCSI devices with their generic names. sg_readcap and sg_format address the /dev/sgN name, not /dev/sdX.

   lsscsi -g

   ReadingThe last column is the /dev/sgN name for each /dev/sdX. A '-' there can mean the sg module is not loaded; load it with modprobe sg and run lsscsi -g again. If a disk you installed is not listed on its own, check the controller mode (path p3).
2. 2.Check what the kernel recorded for size and sector size. Do not trust LOG-SEC alone on a suspect drive.

   lsblk -o NAME,SIZE,LOG-SEC,PHY-SEC

   ReadingLOG-SEC and PHY-SEC are the sizes the kernel recorded, not a fresh query of the drive. When Linux rejects a 520-byte drive it sets the capacity to 0 and falls back to 512, so SIZE 0B is the tell and LOG-SEC can still read 512. Confirm with sg_readcap.
3. 3.Read the kernel's own verdict on the drive's format.

   dmesg | grep -iE 'sector size|DIF|protection'

   Reading'Unsupported sector size 520.' means the drive is formatted with 520-byte blocks. 'Enabling DIF Type 1/2/3 protection' with no 'Disabling' line before it means protection information is in use. A 'Disabling DIF Type N protection' line followed by 'Enabling DIF Type 0 protection' means the drive is protection-formatted but this host cannot use it. 'formatted with unsupported protection type' means a protection type the kernel does not support (above type 3), and the kernel disabled the disk.
4. 4.Ask the drive directly. READ CAPACITY(16) returns the block size and the protection setting.

   sg_readcap --long /dev/sgN

   Reading'Logical block length=520 bytes' is the problem; 512 or 4096 is usable. 'Protection: prot_en=1, p_type=0' with '[type 1 protection]' means protection information is on (the type is p_type+1); prot_en=0 means none. 'Logical blocks per physical block exponent' gives the physical size.
5. 5.Cross-check with smartctl.

   smartctl -i /dev/sgN

   Reading'Logical block size: 520 bytes' confirms the sector size. 'Formatted with type 1 protection' (or type 2 or 3) means protection information is enabled even when the block size reads 512.
6. 6.Record the vendor, model and firmware revision before changing anything, and confirm this is the drive you mean to work on.

   sg_inq /dev/sgN

   ReadingShows the vendor, model and revision from the standard INQUIRY data. Note the array-branded model and the firmware revision before you format.
7. 7.Run sg_format without --format. In this form it only reports and does not touch the disk.

   sg_format /dev/sgN

   ReadingPrints the current block size and block count and does not modify the disk. A TrueNAS community guide shows this output as 'Block size=520' on an affected drive.

1. 1.Before anything else: every command in this path destroys all data on the drive. Copy off anything you need. Run the format inside tmux or on the console, do not reset or power cycle the drive until it finishes, and avoid running other tools against the drive while it formats. An interrupted format can leave the disk 'format corrupt', which takes another format to recover.
2. 2.This destroys all data on the drive. Change the logical block size to 512. sg_format sends MODE SELECT to set the size, then FORMAT UNIT. With the default --fmtpinfo=0 it formats with protection type 0, meaning no protection information.
   Destructive. This command destroys what is on the device and cannot be undone. Confirm the device path twice and have a backup of anything you still need before running it.

   sg_format --format --size=512 /dev/sgN

   Readingsg_format polls the drive every 60 seconds and prints a percentage. A full format can take many hours, and days on hard disks over 10 TB.
3. 3.This also destroys all data on the drive. Use it instead of the 512 command if you want 4096-byte logical blocks: some SCSI disks accept 4096 as an alternate size, and which sizes a drive accepts depends on the model. Same timing and interruption rules as above.
   Destructive. This command destroys what is on the device and cannot be undone. Confirm the device path twice and have a backup of anything you still need before running it.

   sg_format --format --size=4096 /dev/sgN

   ReadingProgress and timing as for the 512 format. If the drive rejects 4096, it may still accept 512.
4. 4.If sg_readcap showed 512 bytes with prot_en=1, the block size is already right and protection information is what the host objects to. The 512 command above formats with protection type 0 by default, so it also clears protection, and it destroys all data in the same way. One TrueNAS community guide ran protection removal and the size change as two separate formats; the man page does not require that.
5. 5.When the format finishes, read the capacity again to confirm the result.

   sg_readcap --long /dev/sgN

   Reading'Logical block length=512 bytes' (or 4096) and 'prot_en=0' means the drive is in a format Linux accepts. Anything else after an interrupted run means the format did not complete; run it again.

1. 1.This destroys all data on the drive. If the default command is rejected, retry with --six, which uses 6-byte MODE SENSE and MODE SELECT instead of the default 10-byte commands. One Level1Techs poster with NetApp drives in a DS2246 shelf behind a Dell H200e reported needing it; that is a forum report, not a documented brand rule. Separately, in the same thread, the poster who hit 'Parameter list length error' on the first attempt reported fixing it by reissuing the format at 512 immediately after a format at 520; that is a single report.
   Destructive. This command destroys what is on the device and cannot be undone. Confirm the device path twice and have a backup of anything you still need before running it.

   sg_format --format --size=512 --six /dev/sgN

   ReadingSame progress output, timing and interruption rules as the default format.
2. 2.This destroys all data on the drive. On Seagate drives, Seagate documents changing the sector size between 512 and 4096 with SeaChest_Format on models that support it. Check the model's data sheet, manual or SeaChest's Feature Supported output first, and disable out-of-band management services, smartd among them, before you start. Seagate notes that a sector-size change may affect every LUN or namespace on a multi-LUN device. Seagate's page does not mention 520-byte drives, so treat this as a second route, not a known fix.
   Destructive. This command destroys what is on the device and cannot be undone. Confirm the device path twice and have a backup of anything you still need before running it.

   SeaChest_Format -d /dev/sgN --setSectorSize 4096 --confirm this-will-erase-data-and-may-render-the-drive-inoperable

   ReadingSeagate says the longest wait for a SATA drive is 5 hours, and that any interruption while formatting may render the device inoperable.
3. 3.After the SeaChest change, check the drive with SeaChest's information option, as Seagate's page shows.

   SeaChest_Format -d /dev/sgN -i

   ReadingCheck the sector size the drive now reports. sg_readcap --long gives the same check.
4. 4.Do not repeat --format three times to force a size change. The man page says three or more --format flags skip MODE SENSE and MODE SELECT, and in that mode sg_format cannot change the logical block size. One forum reply reported otherwise; it is a single report against the documentation.
5. 5.Check what the drive is connected to. Where a thread named the controller and its mode, it was an IT-mode card or an HBA (Dell H200 and H710 in IT mode, an LSI 9200-8e in IT mode, an HPE H240); one Level1Techs poster used a Dell PERC200 and did not say which mode. TrueNAS warns that hardware RAID controllers can mask disk serial numbers and SMART data and recommends HBA mode or IT-mode firmware. We found no vendor document stating that a RAID-mode controller blocks the format, but a disk behind a RAID-mode controller may not be listed individually, so if the disk has no /dev/sgN entry of its own, the controller mode is the first thing to check.
6. 6.Know when to stop. An interrupted format can leave the disk 'format corrupt', and another format is the remedy. In the threads we read, drives that returned Illegal Request or Hardware Error to every attempt were never reformatted: a NetApp drive returning Illegal Request to both sg_format and setblocksize, ex-NetApp Seagate ST3000NM0023 drives that refused every option, and a NetApp Samsung PM1643 SAS SSD that failed with 'Device internal reset'. One TrueNAS user's format stalled at 89.99% after three days, twice; replies there raised drive heat, not a diagnosis. For a drive that refuses every attempt, a replacement drive is the remaining option.

The sg_format man page says a full format can take many hours, and days on hard disks over 10 TB. In the threads, one Level1Techs reply reported anywhere from 10 minutes to 10 hours, and one TrueNAS user saw a format stall at 89.99% after three days, twice; replies there warned about drive heat during long formats.

Not in any thread we found. A TrueNAS user with over 100 EMC-sourced SATA SSDs could not reformat them with sg_format, a reply said 'sg_format only works with sas/scsi disks', and no fix for the SATA drives was recorded. A NetApp Samsung SM1633 SAS SSD in the same thread did reformat with sg_format. This rests on forum reports only.

We found no document that says so. In one ServeTheHome thread a NetApp-branded Samsung PM1643 failed to reformat after a firmware update from NA52 to NA54, and a reply noted an earlier success was on NA53 and suggested NA54 may have removed the option. Record the firmware revision with sg_inq before changing anything.