HBA or RAID controller for TrueNAS: can you keep the card you have?

1. 1.Identify the controller and the kernel driver bound to it, from a Linux shell on the server. -k shows the driver in use for each device and -nn shows vendor and device codes as both numbers and names.

   lspci -nnk | grep -i -A3 -E 'sas|raid'

   ReadingThe 'Kernel driver in use:' line names the driver. mpt3sas is the Linux driver for PCI-Express 12Gb/s SAS host adapters; hpsa is the driver for HP Smart Array controllers. Note the name whatever it is: the lsblk and smartctl tests that follow decide whether the drives behind it reach the OS one by one.
2. 2.List the block devices with their model and serial. ZFS needs one device per physical drive, carrying that drive's own identity. In one TrueNAS thread a T430 owner whose PERC H730 was set up without RAID saw no disks at all in TrueNAS after a drive upgrade.

   lsblk -o NAME,MODEL,SERIAL,SIZE,TRAN

   ReadingExpect one line per physical drive, with the drive's own model and the serial printed on its label. One large device instead of several, a model string naming the controller rather than the drive, or serials that do not match the labels mean the card is presenting virtual disks; TrueNAS warns that RAID cards can mask disk serial numbers. If a column is missing, run lsblk --list-columns to see what this lsblk version offers.
3. 3.Ask smartmontools which devices it finds.

   smartctl --scan

   ReadingPrints each device name, device type and protocol. This list does not decide the question on its own: a '-d scsi' entry does not by itself show that the OS reaches the physical drive directly. Decide on the lsblk model and serial check above and the smartctl -a test in path p2.

1. 1.Read SMART from each drive the plain way, with no device type. This is the test that matters most: TrueNAS warns that some repurposed RAID cards mask SMART health information.

   smartctl -a /dev/sdX

   ReadingIf the 'Device Model' or 'Product' line and the 'Serial Number' line show the physical drive and the SMART attributes or health section is populated, SMART passes through. If the command errors, or identifies a controller virtual disk with no drive attributes, the OS cannot read that drive's SMART directly.
2. 2.If the plain read fails on a MegaRAID or Dell PERC card, read the drive through the controller instead. smartmontools says this device type also works for Dell PERC controllers; N (0-127) selects the disk on the controller.

   smartctl -a -d megaraid,N /dev/sdX

   ReadingFull SMART data here where the plain read failed means the drive is reachable only through the controller's RAID layer, so ZFS is still not seeing the disk itself.
3. 3.On an HP Smart Array controller (hpsa driver), the equivalent device type is cciss. A DL380 Gen9 owner with a Smart Array P840 asked the same keep-or-replace question on the TrueNAS forums and was waiting on an LSI HBA.

   smartctl -a -d cciss,N /dev/sgN

   Readingsmartmontools' example is 'smartctl -a -d cciss,0 /dev/sg2'. Data returned only this way means the disk sits behind the Smart Array layer.

1. 1.List each vdev and its member devices. OpenZFS says hardware RAID limits ZFS's ability to self-heal on checksum failures, because ZFS cannot treat the disk holding a bad sector as bad when the controller handles the redundancy, unless ZFS stores a duplicate copy. In one TrueNAS thread, a pool built on a single PERC H730P RAID5 virtual disk went SUSPENDED after I/O errors while the PERC BIOS showed every drive Online.

   zpool status -v

   ReadingA vdev whose only member is one controller virtual disk (for example one RAID5 virtual disk) holds no ZFS redundancy of its own across disks, which is the layout OpenZFS says limits ZFS's ability to self-heal on checksum failures.
2. 2.If the pool sits on a controller virtual disk, copy the data to other storage and check the copy before you change the card. Moving a PERC 9 to HBA mode requires deleting every virtual disk on it (path p4), which destroys what is on them.

1. 1.Confirm the card is one Dell's guide covers. The PERC 9 user guide names the H330, H730 and H830 and documents two personality modes: RAID mode, the usual factory setting, and HBA mode, in which the card creates no virtual disks and every physical disk is a non-RAID disk under operating-system control.
2. 2.Data loss: this step deletes every virtual disk on the card, and the data on those arrays is lost. Copy everything off the arrays and check the copy before you start. Then prepare the card: Dell lists these prerequisites for the RAID to HBA transition: all virtual disks removed or deleted, hot spares removed or re-purposed, foreign configurations cleared, failed disks removed and any local SED security key deleted.
   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.
3. 3.Data loss: this switch ends RAID mode on the card, and Dell requires its virtual disks to be deleted first, so no array on it survives. Do it only after the previous step, with the data copied off and checked. Open the Dell PERC 9 Configuration Utility (HII) and use the menu path below, then reboot.
   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.

   Dell PERC 9 Configuration Utility (HII): Controller Management > Advanced Controller Management > Switch to HBA mode

   ReadingThe new personality takes effect after the reboot.
4. 4.After the reboot, run lsblk and the plain smartctl read from paths p1 and p2 again. Dell states that in HBA mode SMART monitoring is disabled and the controller does not report SMART errors. That sentence is about the controller's own monitoring: no Dell page we read says whether the OS can still read each drive's SMART data through the card, so this test decides it. Forum replies disagree on the point for the H730 Mini, which is why we do not take either side.

   smartctl -a /dev/sdX

   ReadingEach drive's own model, serial and a populated SMART section mean TrueNAS can monitor the drive. An error, or a device with no drive attributes, means it cannot: go to path p5.

1. 1.Know that the two primary sources disagree on the fallback. TrueNAS says that where a RAID card has no HBA mode, a RAID0 can be configured for every disk. OpenZFS says single-drive RAID0 arrays are not recommended, for many of the same reasons as other hardware RAID. We print both and do not pick one.
2. 2.If you take the RAID0-per-disk route, know what the cache does. On a PERC 9 write-back is the default virtual-disk cache policy, and Dell says it is also supported on single-drive RAID0 virtual disks; Dell also says all RAID volumes are presented to Windows and Linux as write-through whatever the virtual disk's actual policy. A virtual disk can be forced into write-back with no battery present; Dell recommends a power backup system when forcing it. OpenZFS says a RAID write cache adds a failure point, caches all writes rather than only synchronous ones, and can lose data if the battery fails when it is needed during a power loss.
3. 3.Otherwise replace the card. TrueNAS's guide names Broadcom (formerly Avago, formerly LSI) SAS host bus adapters flashed with the latest IT or Target Mode firmware, and OpenZFS says an HBA is best for both performance and reliability. TrueNAS also says not to use a motherboard's built-in hardware RAID. What IT firmware is, and which cards need a flash, is covered on the IT mode page linked below.

TrueNAS says a card in HBA mode (also called passthrough or JBOD mode) performs indistinguishably from a standard HBA. Community threads contest that for specific cards: replies in TrueNAS and Level1Techs threads called PERC passthrough quirky, a data-loss risk or, citing other users' reports, a mode that hides SMART, while another reply called those concerns FUD, and a long-standing TrueNAS community explainer says RAID controllers generally hide the physical device even in JBOD or HBA mode. That is forum opinion, not a specification, so test your own card with lsblk and smartctl rather than trusting either side.

No. In RAID mode a PERC 9 can convert individual RAID-capable disks to non-RAID disks from the BIOS configuration utility or the UEFI/HII RAID configuration utility, which is separate from switching the whole controller's personality. In HBA mode the card creates no virtual disks at all and every disk is a non-RAID disk under operating-system control. Whichever you use, run smartctl -a /dev/sdX on each drive afterwards.

The Dell guide we read is the PERC 9 guide, which covers the H330, H730 and H830. We have not read a Dell document on the H710 or H740, or an HPE document on Smart Array HBA mode, so this page states nothing about their modes. The tests here work on any card: if lsblk lists each drive with its own serial and smartctl -a /dev/sdX reads its SMART data without -d megaraid,N or -d cciss,N, the OS reaches the disk directly.

OpenZFS says a failed RAID controller can require replacement with the same model. In one TrueNAS thread, pools exported from an R740xd with a PERC H740 did not appear on a second R740xd with a PERC H730; with no HBA330 to hand, the poster planned a fresh install and an rsync instead.